{ "conference": "NeurIPS 2026", "source": "OpenReview NeurIPS.cc/2026/Workshop + workshop websites", "count": 104, "groups": [ { "id": "fm", "label": "Foundation Models & LLMs", "count": 13 }, { "id": "agents", "label": "Agents & Agentic Systems", "count": 24 }, { "id": "science", "label": "AI for Science", "count": 19 }, { "id": "health", "label": "Health & Medicine", "count": 7 }, { "id": "trust", "label": "Trustworthy & Safe AI", "count": 14 }, { "id": "theory", "label": "Theory & Methods", "count": 11 }, { "id": "society", "label": "Society, Applications & Community", "count": 16 } ], "workshops": [ { "key": "HAIC", "title": "2026 NeurIPS Workshop on Human-AI Coevolution (HAIC)", "subtitle": "Measuring human-agent teams in the agentic era", "summary": "HAIC 2026 builds a methodological foundation for the empirical evaluation of human-agent teams, asking how to evaluate and govern human-agent systems rigorously as they coevolve with their users, grounded in case studies from healthcare, mental health, aviation, and finance.", "cfp_full": "HAIC @ NeurIPS 2026 - Human-AI Coevolution\nMeasuring human-agent teams in the agentic era. Second edition of the HAIC workshop series.\n\nAbout the workshop - Evaluation has not kept pace with deployment\n\nCoding agents handle substantial portions of professional software workflows, clinical decision-support agents triage patients, and conversational agents mediate human learning and relationships. The community's evaluation methods, built largely for static benchmark performance on chat completion, have not adapted at the same pace.\n\nA recent review of agentic-AI evaluation found that 83% of evaluations are dominated by technical metrics, while human-centered (30%), safety (53%), and economic dimensions (30%) remain peripheral (Jafari Meimandi et al., 2025). HAIC 2026 builds a methodological foundation for the empirical evaluation of human-agent teams. The central question is how to evaluate and govern human-agent systems rigorously as they coevolve with the people who use them, both in general deployment and in domains where the gap between benchmark and reality has the highest stakes?\n\nThis is the second edition of the HAIC series, following the inaugural ICLR 2025 workshop on human-AI coevolution. Where the first workshop mapped coevolution broadly across five themes, this one commits to a single focused operationalization: rigorous empirical evaluation of human-agent teams, with each theme anchored in recent peer-reviewed evidence.\n\nThe program grounds its discussion in case studies from high-stakes domains where the organizing team has direct research access: healthcare, mental health, aviation, and finance. The format weights discussion over talks, with breakouts feeding an open-problems registry and a community position paper.\n\nThemes\n\nThe themes are not independent. Deployment moves the validity target (Theme 1), the human feedback meant to correct course is itself contested (Theme 2), and evaluation must adapt as systems and users coevolve (Theme 3).\n\nTheme 1 - Validity of evaluation in deployed contexts: Benchmarks assume a static target. In deployment the target moves, and the constructs being measured, such as productivity, helpfulness, and safety, are themselves contested across domains. How do validity frameworks adapt?\n\nTheme 2 - Expert disagreement and the limits of human feedback: RLHF assumes aggregated feedback approximates a coherent target. A growing line of work treats disagreement as signal rather than noise. When does it mark evaluation invalidity, and when does it reflect domain pluralism that deployed systems should preserve?\n\nTheme 3 - Adaptive testing and continual evaluation: Deployed teams coevolve: skills reallocate, populations shift, distributions drift. A fixed test set can lose validity with no visible signal. We seek methods that track and respond to drift.\n\nThe three themes form a cycle: Theme 1 feeds Theme 2, Theme 2 feeds Theme 3, and Theme 3 returns to Theme 1 as drift moves the evaluation target again.\n\nCall for papers - Scope of submissions\n\nWe invite new evaluation methods and benchmarks, datasets of human-agent interaction, reproducible benchmark critiques, case studies from high-stakes domains, and position papers. The workshop is non-archival, and each submission receives three double-blind reviews via OpenReview.\n\nKey dates (Dates follow the NeurIPS 2026 recommended workshop timeline. All deadlines are Anywhere on Earth.)\n- Submission deadline: August 29, 2026 (AoE) - Double-blind via OpenReview\n- Acceptance notification: September 29, 2026 (AoE) - Hard deadline\n- Camera-ready: To be announced - Set with authors after decisions\n- Workshop, Atlanta: December 12-13, 2026 - One day within that window, confirmed by NeurIPS", "cfp_status": "published", "topics": [ "Validity of evaluation in deployed contexts", "Expert disagreement and the limits of human feedback", "Adaptive testing and continual evaluation", "New evaluation methods and benchmarks", "Datasets of human-agent interaction", "Reproducible benchmark critiques", "Case studies from high-stakes domains (healthcare, mental health, aviation, finance)", "Position papers" ], "important_dates": [ { "label": "Submission deadline (AoE)", "date": "2026-08-29" }, { "label": "Acceptance notification (AoE)", "date": "2026-09-29" }, { "label": "Camera-ready", "date": "TBA" }, { "label": "Workshop (Atlanta)", "date": "December 12-13, 2026" } ], "organizers": [ "Kiana Jafari (General chair, Stanford University)", "Marc Schlichting (Submissions & review, Stanford University)", "Dylan Asmar (Program committee, MIT Lincoln Laboratory)", "Ahmad Rushdi (Sponsorship & inclusion, Stanford University (HAI))", "Martin Gonzalez (Outreach, Google DeepMind)" ], "speakers": [], "host_url": "http://Neurips2026haic.com", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/HAIC", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/HAIC", "location": "Atlanta, GA, United States", "city": "Atlanta", "workshop_date": "", "contact": "neurips2026haic@gmail.com", "tracks": [ { "key": "HAIC", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/HAIC", "submission_dates_raw": "Submission Deadline: Aug 30 2026 11:59AM UTC-0" } ], "group": "agents", "group_label": "Agents & Agentic Systems", "corpus": "2026 NeurIPS Workshop on Human-AI Coevolution (HAIC) Measuring human-agent teams in the agentic era HAIC 2026 builds a methodological foundation for the empirical evaluation of human-agent teams, asking how to evaluate and govern human-agent systems rigorously as they coevolve with their users, grounded in case studies from healthcare, mental health, aviation, and finance. Validity of evaluation in deployed contexts Expert disagreement and the limits of human feedback Adaptive testing and continual evaluation New evaluation methods and benchmarks Datasets of human-agent interaction Reproducible benchmark critiques Case studies from high-stakes domains (healthcare, mental health, aviation, finance) Position papers HAIC @ NeurIPS 2026 - Human-AI Coevolution\nMeasuring human-agent teams in the agentic era. Second edition of the HAIC workshop series.\n\nAbout the workshop - Evaluation has not kept pace with deployment\n\nCoding agents handle substantial portions of professional software workflows, clinical decision-support agents triage patients, and conversational agents mediate human learning and relationships. The community's evaluation methods, built largely for static benchmark performance on chat completion, have not adapted at the same pace.\n\nA recent review of agentic-AI evaluation found that 83% of evaluations are dominated by technical metrics, while human-centered (30%), safety (53%), and economic dimensions (30%) remain peripheral (Jafari Meimandi et al., 2025). HAIC 2026 builds a methodological foundation for the empirical evaluation of human-agent teams. The central question is how to evaluate and govern human-agent systems rigorously as they coevolve with the people who use them, both in general deployment and in domains where the gap between benchmark and reality has the highest stakes?\n\nThis is the second edition of the HAIC series, following the inaugural ICLR 2025 workshop on human-AI coevolution. Where the first workshop mapped coevolution broadly across five themes, this one commits to a single focused operationalization: rigorous empirical evaluation of human-agent teams, with each theme anchored in recent peer-reviewed evidence.\n\nThe program grounds its discussion in case studies from high-stakes domains where the organizing team has direct research access: healthcare, mental health, aviation, and finance. The format weights discussion over talks, with breakouts feeding an open-problems registry and a community position paper.\n\nThemes\n\nThe themes are not independent. Deployment moves the validity target (Theme 1), the human feedback meant to correct course is itself contested (Theme 2), and evaluation must adapt as systems and users coevolve (Theme 3).\n\nTheme 1 - Validity of evaluation in deployed contexts: Benchmarks assume a static target. In deployment the target moves, and the constructs being measured, such as productivity, helpfulness, and safety, are themselves contested across domains. How do validity frameworks adapt?\n\nTheme 2 - Expert disagreement and the limits of human feedback: RLHF assumes aggregated feedback approximates a coherent target. A growing line of work treats disagreement as signal rather than noise. When does it mark evaluation invalidity, and when does it reflect domain pluralism that deployed systems should preserve?\n\nTheme 3 - Adaptive testing and continual evaluation: Deployed teams coevolve: skills reallocate, populations shift, distributions drift. A fixed test set can lose validity with no visible signal. We seek methods that track and respond to drift.\n\nThe three themes form a cycle: Theme 1 feeds Theme 2, Theme 2 feeds Theme 3, and Theme 3 returns to Theme 1 as drift moves the evaluation target again.\n\nCall for papers - Scope of submissions\n\nWe invite new evaluation methods and benchmarks, datasets of human-agent interaction, reproducible benchmark critiques, case studies from high-stakes domains, and position papers. The workshop is non-archival, and each submission receives three double-blind reviews via OpenReview.\n\nKey dates (Dates follow the NeurIPS 2026 recommended workshop timeline. All deadlines are Anywhere on Earth.)\n- Submission deadline: August 29, 2026 (AoE) - Double-blind via OpenReview\n- Acceptance notification: September 29, 2026 (AoE) - Hard deadline\n- Camera-ready: To be announced - Set with authors after decisions\n- Workshop, Atlanta: December 12-13, 2026 - One day within that window, confirmed by NeurIPS" }, { "key": "AABA4ET", "title": "2nd Workshop on Agentic AI Benchmarks and Applications for Enterprise Tasks (NeurIPS 2026)", "subtitle": "AABA4ET NeurIPS 2026", "summary": "The 2nd edition (after AAAI 2026) of a workshop focused on benchmarking, evaluating, and deploying agentic AI systems for complex, dynamic enterprise operations, bridging cutting-edge agentic AI research with the practical demands of real-world enterprise deployment.", "cfp_full": "2nd Workshop on Agentic AI Benchmarks and Applications for Enterprise Tasks\nDecember 11 or 12, 2026 | Sydney, Australia\n\nWhere Agentic AI Meets the Real World of Work — Benchmarking, Evaluating, and Deploying Intelligent Agents for Complex Enterprise Operations at Scale.\n\nThe primary goal of this workshop is to foster discussions and collaborations to build robust, efficient, and trustworthy Agentic AI technologies for complex and dynamic enterprise business operations. It aims to bridge the gap between cutting-edge Agentic AI research and the practical demands of enterprise deployment and rigorous evaluation.\n\nThis is the 2nd edition of the workshop, following the success of the 1st Workshop on Agentic AI Benchmarks and Applications for Enterprise Tasks, held at AAAI 2026 in Singapore.\n\nThe workshop will address the following specific issues:\n- Benchmarking and Evaluation: Addressing the urgent need for robust benchmarks, datasets, and metrics to reliably evaluate Agentic AI systems for enterprise-level performance, safety, and reliability, including the challenges of creating realistic and representative enterprise task environments.\n- Application of Agentic AI in Enterprise Settings: Exploring how Agentic AI can perform complex tasks such as understanding on-site operations, planning, observation, reflection, and system management within enterprise contexts.\n- Safety, Robustness, and Trustworthiness in Real-World Deployment: failure detection and recovery, robustness to distribution shift and adversarial/unexpected inputs, guardrails and oversight mechanisms, and building trust in long-running autonomous agents in production.\n- Human-Agent Interaction in Enterprise Workflows: Development and deployment of intelligent assistants that augment human capabilities in business operations.\n- Multimodal Reasoning for Enterprise Tasks: Integrating multimodal LLMs to handle diverse physical data (e.g., visual, textual, auditory) for robust decision-making and task execution in enterprises.\n- Task Planning and Orchestration in Enterprise Environment: Strategies for integrating multiple agents and tools to achieve complex, multi-step enterprise goals.\n- Others: We are also seeking a wide range of content related to Agentic AI technology.\n\nSubmission Guidelines\nPapers must use the official NeurIPS 2026 LaTeX style file and are limited to up to 4 pages, including all figures and tables; references and appendices do not count toward this limit. Submissions require double-blind review, with author names, affiliations, and any other identifying information removed at submission time. Submit a single anonymized PDF (main paper + references + optional appendix combined). The workshop welcomes work that is under review or to be submitted to other venues. Submissions are non-archived and may be published on non-archived submission servers such as arXiv. For accepted papers, at least one author must attend the workshop in person to present the work as a poster or oral presentation.\n\nImportant Dates\n- Submission Start: August 2, 2026 (AoE)\n- Submission Deadline: August 30, 2026 (AoE)\n- Acceptance Notification: September 29, 2026 (AoE)\n- Workshop Date: December 11 or 12, 2026 (Sydney, Australia)", "cfp_status": "published", "topics": [ "Benchmarking and Evaluation of Agentic AI for enterprise tasks", "Application of Agentic AI in Enterprise Settings", "Safety, Robustness, and Trustworthiness in Real-World Deployment", "Human-Agent Interaction in Enterprise Workflows", "Multimodal Reasoning for Enterprise Tasks", "Task Planning and Orchestration in Enterprise Environment", "Other Agentic AI technology topics" ], "important_dates": [ { "label": "Submission Start", "date": "2026-08-02" }, { "label": "Submission Deadline", "date": "2026-08-30" }, { "label": "Acceptance Notification", "date": "2026-09-29" }, { "label": "Workshop Date", "date": "2026-12-11 or 2026-12-12" } ], "organizers": [], "speakers": [], "host_url": "https://sites.google.com/view/aaba4et", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AABA4ET", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AABA4ET", "location": "Sydney, Australia", "city": "Sydney", "workshop_date": "", "contact": "neurips26ws-aaba4et-pc@googlegroups.com", "tracks": [ { "key": "AABA4ET", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AABA4ET", "submission_dates_raw": "Submission Start: Aug 02 2026 12:00PM UTC-0, Submission Deadline: Aug 30 2026 12:00PM UTC-0" } ], "group": "agents", "group_label": "Agents & Agentic Systems", "corpus": "2nd Workshop on Agentic AI Benchmarks and Applications for Enterprise Tasks (NeurIPS 2026) AABA4ET NeurIPS 2026 The 2nd edition (after AAAI 2026) of a workshop focused on benchmarking, evaluating, and deploying agentic AI systems for complex, dynamic enterprise operations, bridging cutting-edge agentic AI research with the practical demands of real-world enterprise deployment. Benchmarking and Evaluation of Agentic AI for enterprise tasks Application of Agentic AI in Enterprise Settings Safety, Robustness, and Trustworthiness in Real-World Deployment Human-Agent Interaction in Enterprise Workflows Multimodal Reasoning for Enterprise Tasks Task Planning and Orchestration in Enterprise Environment Other Agentic AI technology topics 2nd Workshop on Agentic AI Benchmarks and Applications for Enterprise Tasks\nDecember 11 or 12, 2026 | Sydney, Australia\n\nWhere Agentic AI Meets the Real World of Work — Benchmarking, Evaluating, and Deploying Intelligent Agents for Complex Enterprise Operations at Scale.\n\nThe primary goal of this workshop is to foster discussions and collaborations to build robust, efficient, and trustworthy Agentic AI technologies for complex and dynamic enterprise business operations. It aims to bridge the gap between cutting-edge Agentic AI research and the practical demands of enterprise deployment and rigorous evaluation.\n\nThis is the 2nd edition of the workshop, following the success of the 1st Workshop on Agentic AI Benchmarks and Applications for Enterprise Tasks, held at AAAI 2026 in Singapore.\n\nThe workshop will address the following specific issues:\n- Benchmarking and Evaluation: Addressing the urgent need for robust benchmarks, datasets, and metrics to reliably evaluate Agentic AI systems for enterprise-level performance, safety, and reliability, including the challenges of creating realistic and representative enterprise task environments.\n- Application of Agentic AI in Enterprise Settings: Exploring how Agentic AI can perform complex tasks such as understanding on-site operations, planning, observation, reflection, and system management within enterprise contexts.\n- Safety, Robustness, and Trustworthiness in Real-World Deployment: failure detection and recovery, robustness to distribution shift and adversarial/unexpected inputs, guardrails and oversight mechanisms, and building trust in long-running autonomous agents in production.\n- Human-Agent Interaction in Enterprise Workflows: Development and deployment of intelligent assistants that augment human capabilities in business operations.\n- Multimodal Reasoning for Enterprise Tasks: Integrating multimodal LLMs to handle diverse physical data (e.g., visual, textual, auditory) for robust decision-making and task execution in enterprises.\n- Task Planning and Orchestration in Enterprise Environment: Strategies for integrating multiple agents and tools to achieve complex, multi-step enterprise goals.\n- Others: We are also seeking a wide range of content related to Agentic AI technology.\n\nSubmission Guidelines\nPapers must use the official NeurIPS 2026 LaTeX style file and are limited to up to 4 pages, including all figures and tables; references and appendices do not count toward this limit. Submissions require double-blind review, with author names, affiliations, and any other identifying information removed at submission time. Submit a single anonymized PDF (main paper + references + optional appendix combined). The workshop welcomes work that is under review or to be submitted to other venues. Submissions are non-archived and may be published on non-archived submission servers such as arXiv. For accepted papers, at least one author must attend the workshop in person to present the work as a poster or oral presentation.\n\nImportant Dates\n- Submission Start: August 2, 2026 (AoE)\n- Submission Deadline: August 30, 2026 (AoE)\n- Acceptance Notification: September 29, 2026 (AoE)\n- Workshop Date: December 11 or 12, 2026 (Sydney, Australia)" }, { "key": "AutoMLR", "title": "40th Conference on Neural Information Processing Systems Workshop for Autonomous Machine Learning Research", "subtitle": "NeurIPS 2026 AutoMLR Workshop", "summary": "The Workshop for Autonomous Machine Learning Research (AutoMLR) is a discussant-led, non-archival NeurIPS 2026 workshop where ML research produced or decisively advanced by autonomous agents is evaluated through independent review and public author-discussant dialogue, keeping human judgment central.", "cfp_full": "Workshop for Autonomous Machine Learning Research\nNeurIPS 2026 - Sydney - In person\nWorkshop date: Dec 11 or Dec 12\nFormat: Non-archival - Discussant-led\n\nAutonomous research, human judgment. A discussant-led workshop where autonomous ML research is evaluated through independent review and public dialogue.\n\n01 - Do conferences exist for science, or for scientists?\nThe research process comprises the formulation of a hypothesis, the design of an experiment, and the judgment of a result. The implicit assumption that this process is a fundamentally human act is suddenly being challenged by autonomous research.\nGiven the development of agent harnesses that can pursue long-horizon tasks, we believe it is time for the community to formally recognize and plan for the inevitability of impactful autonomous research.\nOur goal is to make autonomous research meaningful in a way that strengthens the ML community without undermining the role of human researchers. The workshop brings this structural change into the open and anchors autonomous research in human judgment and participation. Each accepted paper is presented in person by an author. A qualified non-author serves as its discussant, responding to the work and helping lead discussion with the audience. This format grounds autonomous research in the human work of communicating, scrutinizing, and building shared scientific understanding.\nJust as Pandora's Box cannot be closed, the effect of frontier models on researchers now and in the future cannot be undone.\nHowever, just as Hope was discovered at the bottom of the box, we believe the normalization of autonomous research will foster an environment where scientists can engage in open, constructive dialogue about AI-generated science.\n\n02 - Paper format and submission\nEligibility\nWe invite machine learning research in which an autonomous agent either conducted the research end-to-end or made a decisive contribution to the paper's primary result. The paper must state the qualifying result in its abstract. Without the agent's contribution, the qualifying result could not have been established, or the paper's main conclusions would be materially different.\nEvery submission must include both a qualifying ML research result and a detailed account of the autonomous system that produced it. Neither component is sufficient on its own. Autonomous research in other scientific domains is outside the workshop's scope.\n\nRequired paper structure\nYour paper must have three clearly labelled parts. Parts 1 and 2 each have a four-page maximum; Part 3 has a one-page maximum. Prepare your submission using the Overleaf template.\nPart 1 - 4 pages - Auto research result: Present the machine learning research itself: the hypothesis, method, evidence, and primary result generated or developed by the agent.\nPart 2 - 4 pages - System design: Describe the agent, harness, tools, research loop, etc. Provide commentary on the significance of the main result.\nPart 3 - 1 page - Reflections: Given autonomous research is a brand, we invite broader reflections on how the field should adapt. We believe this workshop is only a start and hope to facilitate discussion based on these reflections in the Town Hall.\n\nDisclosure policy\nAuthorship remains exclusively human. Authors curate the work, verify its claims, disclose agent involvement, and remain responsible for the final submission.\nBecause this workshop's premise is that an autonomous agent may drive research, we expect much of Part 1 to be generated and written by the agent (including writing, figures, etc.). However, in keeping with wider NeurIPS policy, authors are ultimately responsible for the entire content of the paper, including all text, figures, and references.\nTo ensure that authors remain in compliance with NeurIPS policy, Part 2 of the submission will be an extended meta-analysis and discussion of the role and design of the agent in the research process. We expect Part 2 to be human-written; this will include details on the agent, prompts, harness, tools, research loop, human interventions, and verification.\nAuthors are responsible for ensuring that all content is correct and original and for verifying tool outputs - including guarding against hallucinated results, figures, or citations. Scientific integrity rests with the human authors.\nFinal, exact disclosure instructions will ship with the submission portal; the above is provisional and consistent with the NeurIPS policy on the use of agents and large language models.\n\n03 - Discussant format\nA discussant is a non-author who prepares an independent reading of an accepted paper, responds to the author's summary, and helps lead the ensuing discussion with the audience. The model is common in the social sciences, but remains unfamiliar in machine learning.\nTraditionally, accepted conference papers are presented by their authors, while reviewers disappear into the background. For autonomous research, we propose that the allocation of attention be reversed. If autonomous systems can produce candidate papers at scale, the scarce human contribution becomes the ability to evaluate AI-generated claims.\nThe discussant is not a co-author and does not replace the author. They bring an independent, informed perspective to the work: why its primary result deserves attention, which evidence is most persuasive, what limitations remain, and what the community should discuss next.\nFrom blind review to public dialogue:\n01 - Review - Blind decision-making: Submissions remain blind and reviewers anonymous. Reviewers assess the work independently under the NeurIPS conflict-of-interest policy.\n02 - Selection - An attending discussant: We first invite an accepting reviewer who plans to attend in person. If none is available, the organizers appoint a qualified discussant from the workshop's in-person attendees.\n03 - Preparation - An independent reading: The discussant prepares an account of the contribution, strongest evidence, unresolved limitations, and questions for the author.\n04 - Workshop - Summary, response, dialogue: An attending author briefly presents the work. The discussant responds, opening a dialogue with the author and audience.\nFull review and attribution policy: Each submission must designate min(3, number of authors) authors who agree to review other workshop submissions. Program chairs will ensure that every submission receives at least three independent reviews, supplementing the reviewer pool when necessary. Blind decision-making remains separate from open post-acceptance discussion. After acceptance, accepted papers will be published with their reviews and reviewers. We will first invite an accepting reviewer who expects to attend in person to serve as discussant. If none is available, the organizers will appoint a qualified discussant from the workshop's in-person attendees. The discussant will be identified and credited for their contribution to the session.\n\n04 - Dates and program\nImportant dates:\n- July 15 - Call for papers\n- August 22 - Abstract deadline\n- August 29 - Final submission\n- September 24 - Decisions\n- September 29 - Discussant notification\n- November 13 - Camera-ready\nExact time zones will be posted with the submission instructions.\n\nSubmission\nSubmissions are now open on OpenReview. Prepare your paper using the workshop template before submitting.\nLength and structure. Submissions must contain three clearly labelled parts. Part 1 may be up to four pages, Part 2 may be up to four pages, and Part 3 may be up to one page. These limits apply separately and may not be reallocated between parts. The total limit is nine pages, excluding references.\nSubmission limit. An individual who is listed as first author on one submission may not appear as an author on any other workshop submission.\nReviewing commitment. Each submission must designate min(3, number of authors) authors who agree to review other workshop submissions. A single-author paper supplies one reviewer, a two-author paper supplies two, and a paper with three or more authors supplies three.\nIn-person participation. All paper presentations and discussant sessions will take place in person. Accepted papers should be represented in Sydney by at least one author, who will briefly present the work and participate in the author-discussant session.\nIs the workshop archival? No. Authors may continue developing and submitting their work elsewhere, subject to those venues' policies.\n\nContact: info@automlr.com", "cfp_status": "published", "topics": [ "Autonomous ML research conducted end-to-end or decisively advanced by an autonomous agent", "Agent/system design: harnesses, tools, and research loops for autonomous research", "Meta-analysis and disclosure of agent involvement in the research process", "Evaluation of AI-generated research claims", "Discussant-led review and public dialogue formats for autonomous research", "Reflections on how the ML field should adapt to autonomous research" ], "important_dates": [ { "label": "Call for papers", "date": "2026-07-15" }, { "label": "Abstract deadline", "date": "2026-08-22" }, { "label": "Final submission", "date": "2026-08-29" }, { "label": "Decisions", "date": "2026-09-24" }, { "label": "Discussant notification", "date": "2026-09-29" }, { "label": "Camera-ready", "date": "2026-11-13" }, { "label": "Workshop", "date": "2026-12-11" } ], "organizers": [ "Arjun Prakash", "Aditya Iyer", "Hamish Ivison", "Jack Liell-Cock", "Amy Greenwald", "Nora Ayanian", "David Tao", "Kevin Wang", "Anna Hakhverdyan", "Stephen Crawford", "Zarif Aziz" ], "speakers": [ "Mengdi Wang (Princeton University)", "Sherry Yang (NYU Courant & Google DeepMind)", "Nik Dawson (Burning Glass Institute)" ], "host_url": "https://automlr.com", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AutoMLR", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AutoMLR", "location": "Sydney, Australia", "city": "Sydney", "workshop_date": "", "contact": "info@automlr.com", "tracks": [ { "key": "AutoMLR", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AutoMLR", "submission_dates_raw": "Submission Start: Jul 15 2026 12:00AM UTC-0, Abstract Registration: Aug 22 2026 12:00AM UTC-0, Submission Deadline: Aug 29 2026 12:00AM UTC-0" } ], "group": "agents", "group_label": "Agents & Agentic Systems", "corpus": "40th Conference on Neural Information Processing Systems Workshop for Autonomous Machine Learning Research NeurIPS 2026 AutoMLR Workshop The Workshop for Autonomous Machine Learning Research (AutoMLR) is a discussant-led, non-archival NeurIPS 2026 workshop where ML research produced or decisively advanced by autonomous agents is evaluated through independent review and public author-discussant dialogue, keeping human judgment central. Autonomous ML research conducted end-to-end or decisively advanced by an autonomous agent Agent/system design: harnesses, tools, and research loops for autonomous research Meta-analysis and disclosure of agent involvement in the research process Evaluation of AI-generated research claims Discussant-led review and public dialogue formats for autonomous research Reflections on how the ML field should adapt to autonomous research Workshop for Autonomous Machine Learning Research\nNeurIPS 2026 - Sydney - In person\nWorkshop date: Dec 11 or Dec 12\nFormat: Non-archival - Discussant-led\n\nAutonomous research, human judgment. A discussant-led workshop where autonomous ML research is evaluated through independent review and public dialogue.\n\n01 - Do conferences exist for science, or for scientists?\nThe research process comprises the formulation of a hypothesis, the design of an experiment, and the judgment of a result. The implicit assumption that this process is a fundamentally human act is suddenly being challenged by autonomous research.\nGiven the development of agent harnesses that can pursue long-horizon tasks, we believe it is time for the community to formally recognize and plan for the inevitability of impactful autonomous research.\nOur goal is to make autonomous research meaningful in a way that strengthens the ML community without undermining the role of human researchers. The workshop brings this structural change into the open and anchors autonomous research in human judgment and participation. Each accepted paper is presented in person by an author. A qualified non-author serves as its discussant, responding to the work and helping lead discussion with the audience. This format grounds autonomous research in the human work of communicating, scrutinizing, and building shared scientific understanding.\nJust as Pandora's Box cannot be closed, the effect of frontier models on researchers now and in the future cannot be undone.\nHowever, just as Hope was discovered at the bottom of the box, we believe the normalization of autonomous research will foster an environment where scientists can engage in open, constructive dialogue about AI-generated science.\n\n02 - Paper format and submission\nEligibility\nWe invite machine learning research in which an autonomous agent either conducted the research end-to-end or made a decisive contribution to the paper's primary result. The paper must state the qualifying result in its abstract. Without the agent's contribution, the qualifying result could not have been established, or the paper's main conclusions would be materially different.\nEvery submission must include both a qualifying ML research result and a detailed account of the autonomous system that produced it. Neither component is sufficient on its own. Autonomous research in other scientific domains is outside the workshop's scope.\n\nRequired paper structure\nYour paper must have three clearly labelled parts. Parts 1 and 2 each have a four-page maximum; Part 3 has a one-page maximum. Prepare your submission using the Overleaf template.\nPart 1 - 4 pages - Auto research result: Present the machine learning research itself: the hypothesis, method, evidence, and primary result generated or developed by the agent.\nPart 2 - 4 pages - System design: Describe the agent, harness, tools, research loop, etc. Provide commentary on the significance of the main result.\nPart 3 - 1 page - Reflections: Given autonomous research is a brand, we invite broader reflections on how the field should adapt. We believe this workshop is only a start and hope to facilitate discussion based on these reflections in the Town Hall.\n\nDisclosure policy\nAuthorship remains exclusively human. Authors curate the work, verify its claims, disclose agent involvement, and remain responsible for the final submission.\nBecause this workshop's premise is that an autonomous agent may drive research, we expect much of Part 1 to be generated and written by the agent (including writing, figures, etc.). However, in keeping with wider NeurIPS policy, authors are ultimately responsible for the entire content of the paper, including all text, figures, and references.\nTo ensure that authors remain in compliance with NeurIPS policy, Part 2 of the submission will be an extended meta-analysis and discussion of the role and design of the agent in the research process. We expect Part 2 to be human-written; this will include details on the agent, prompts, harness, tools, research loop, human interventions, and verification.\nAuthors are responsible for ensuring that all content is correct and original and for verifying tool outputs - including guarding against hallucinated results, figures, or citations. Scientific integrity rests with the human authors.\nFinal, exact disclosure instructions will ship with the submission portal; the above is provisional and consistent with the NeurIPS policy on the use of agents and large language models.\n\n03 - Discussant format\nA discussant is a non-author who prepares an independent reading of an accepted paper, responds to the author's summary, and helps lead the ensuing discussion with the audience. The model is common in the social sciences, but remains unfamiliar in machine learning.\nTraditionally, accepted conference papers are presented by their authors, while reviewers disappear into the background. For autonomous research, we propose that the allocation of attention be reversed. If autonomous systems can produce candidate papers at scale, the scarce human contribution becomes the ability to evaluate AI-generated claims.\nThe discussant is not a co-author and does not replace the author. They bring an independent, informed perspective to the work: why its primary result deserves attention, which evidence is most persuasive, what limitations remain, and what the community should discuss next.\nFrom blind review to public dialogue:\n01 - Review - Blind decision-making: Submissions remain blind and reviewers anonymous. Reviewers assess the work independently under the NeurIPS conflict-of-interest policy.\n02 - Selection - An attending discussant: We first invite an accepting reviewer who plans to attend in person. If none is available, the organizers appoint a qualified discussant from the workshop's in-person attendees.\n03 - Preparation - An independent reading: The discussant prepares an account of the contribution, strongest evidence, unresolved limitations, and questions for the author.\n04 - Workshop - Summary, response, dialogue: An attending author briefly presents the work. The discussant responds, opening a dialogue with the author and audience.\nFull review and attribution policy: Each submission must designate min(3, number of authors) authors who agree to review other workshop submissions. Program chairs will ensure that every submission receives at least three independent reviews, supplementing the reviewer pool when necessary. Blind decision-making remains separate from open post-acceptance discussion. After acceptance, accepted papers will be published with their reviews and reviewers. We will first invite an accepting reviewer who expects to attend in person to serve as discussant. If none is available, the organizers will appoint a qualified discussant from the workshop's in-person attendees. The discussant will be identified and credited for their contribution to the session.\n\n04 - Dates and program\nImportant dates:\n- July 15 - Call for papers\n- August 22 - Abstract deadline\n- August 29 - Final submission\n- September 24 - Decisions\n- September 29 - Discussant notification\n- November 13 - Camera-ready\nExact time zones will be posted with the submission instructions.\n\nSubmission\nSubmissions are now open on OpenReview. Prepare your paper using the workshop template before submitting.\nLength and structure. Submissions must contain three clearly labelled parts. Part 1 may be up to four pages, Part 2 may be up to four pages, and Part 3 may be up to one page. These limits apply separately and may not be reallocated between parts. The total limit is nine pages, excluding references.\nSubmission limit. An individual who is listed as first author on one submission may not appear as an author on any other workshop submission.\nReviewing commitment. Each submission must designate min(3, number of authors) authors who agree to review other workshop submissions. A single-author paper supplies one reviewer, a two-author paper supplies two, and a paper with three or more authors supplies three.\nIn-person participation. All paper presentations and discussant sessions will take place in person. Accepted papers should be represented in Sydney by at least one author, who will briefly present the work and participate in the author-discussant session.\nIs the workshop archival? No. Authors may continue developing and submitting their work elsewhere, subject to those venues' policies.\n\nContact: info@automlr.com" }, { "key": "WRL", "title": "8th Robot Learning Workshop: Is Physical AI Going Zero-Shot?", "subtitle": "WRL@NeurIPS 2026", "summary": "The 8th Robot Learning Workshop at NeurIPS 2026 examines whether Physical AI is truly going zero-shot, exploring how foundation models, scaling laws, and agentic reasoning shift the traditional boundaries of robot learning toward embodied generalization and deployment.", "cfp_full": "8th Robot Learning Workshop @ NeurIPS 2026: Is Physical AI Going Zero-Shot?\n\nIs Physical AI truly going zero-shot? Join the debate on foundation models, robotics, deployment, and embodied generalization.\n\nNews\n\nJuly 13th, 2026 — Submissions are now open! Important Dates: Submission Deadline: August 26, 2026 (11:59 PM AoE) | Decisions Announced: September 29, 2026. Submit via OpenReview.\n\nJuly 13th, 2026 — The Robot Learning Workshop is returning to NeurIPS in 2026 for its 8th edition. See you in Sydney!\n\nAbout\n\nThe year 2025 has seen an unprecedented acceleration in the scale and diversity of foundation models for robotics, with the trend continuing in 2026. With the 8th Robot Learning workshop, returning to its native venue at NeurIPS 2026, we aim to explore a provocative question: Is Physical AI going zero-shot?\n\nDriven by the large-scale training of foundation models across diverse data sources and embodiments, we are witnessing the emergence of agentic approaches that allow robots to reason their way through complex, unseen tasks. We propose to critically examine how these paradigms shift the traditional boundaries of robot learning:\n\n- Are we moving past narrow task-specific fine-tuning toward reasoning-based physical agents?\n- What are the implications of scaling laws, diverse training datasets, and multi-modal models in real-world deployment?\n- Can generalization alone lead us to the strong performance levels required in robotics use cases?\n\nWe seek diverse perspectives from the machine learning and robotics communities—both academia and industry—to map the trajectory of zero-shot physical AI.\n\nImportant Dates:\n- Submission Deadline: August 26, 2026 (11:59 PM AoE)\n- Decisions Announced: September 29, 2026\n\nSubmit via OpenReview.", "cfp_status": "published", "topics": [ "Foundation models for robotics", "Zero-shot and reasoning-based physical agents", "Scaling laws and diverse training datasets", "Multi-modal models for robotics", "Real-world deployment of robot learning", "Embodied generalization", "Agentic approaches for robots" ], "important_dates": [ { "label": "Submissions open", "date": "2026-07-13" }, { "label": "Submission Deadline", "date": "2026-08-26" }, { "label": "Decisions Announced", "date": "2026-09-29" }, { "label": "Workshop", "date": "2026-12-10" } ], "organizers": [ "Andrey Kolobov (Microsoft Research, USA)", "Alex Bewley (Google DeepMind, Switzerland)", "Hamidreza Kasaei (University of Groningen, Netherlands)", "Roberto Calandra (TU Dresden, Germany)", "Johannes V. S. Busch (TU Dresden, Germany)", "Moritz Reuss (NVIDIA, Switzerland)", "Jiaxu Xing (University of Zurich, Switzerland)", "Jen Jen Chung (University of Queensland, Australia)", "Markus Wulfmeier (Nomagic, Switzerland/Poland) — Advisor", "Masha Itkina (Toyota Research Institute, USA) — Advisor" ], "speakers": [ "Weiming Zhi (University of Sydney, Australia)", "Huazhe Xu (Tsinghua University, China)", "Maria Bauza Villalonga (Google DeepMind, UK)", "Michael Milford (QUT, Australia)", "Mengdi Xu (Tsinghua University, China)", "Ingmar Posner (University of Oxford, UK)", "Yadan Luo (University of Queensland, Australia)", "Karl Pertsch (Physical Intelligence, USA)" ], "host_url": "https://www.robot-learning.ml/2026/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/WRL", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/WRL", "location": "Sydney, Australia", "city": "Sydney", "workshop_date": "2026-12-10", "contact": "wrl2026organizers@robot-learning.ml", "tracks": [ { "key": "WRL", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/WRL", "submission_dates_raw": "" } ], "group": "agents", "group_label": "Agents & Agentic Systems", "corpus": "8th Robot Learning Workshop: Is Physical AI Going Zero-Shot? WRL@NeurIPS 2026 The 8th Robot Learning Workshop at NeurIPS 2026 examines whether Physical AI is truly going zero-shot, exploring how foundation models, scaling laws, and agentic reasoning shift the traditional boundaries of robot learning toward embodied generalization and deployment. Foundation models for robotics Zero-shot and reasoning-based physical agents Scaling laws and diverse training datasets Multi-modal models for robotics Real-world deployment of robot learning Embodied generalization Agentic approaches for robots 8th Robot Learning Workshop @ NeurIPS 2026: Is Physical AI Going Zero-Shot?\n\nIs Physical AI truly going zero-shot? Join the debate on foundation models, robotics, deployment, and embodied generalization.\n\nNews\n\nJuly 13th, 2026 — Submissions are now open! Important Dates: Submission Deadline: August 26, 2026 (11:59 PM AoE) | Decisions Announced: September 29, 2026. Submit via OpenReview.\n\nJuly 13th, 2026 — The Robot Learning Workshop is returning to NeurIPS in 2026 for its 8th edition. See you in Sydney!\n\nAbout\n\nThe year 2025 has seen an unprecedented acceleration in the scale and diversity of foundation models for robotics, with the trend continuing in 2026. With the 8th Robot Learning workshop, returning to its native venue at NeurIPS 2026, we aim to explore a provocative question: Is Physical AI going zero-shot?\n\nDriven by the large-scale training of foundation models across diverse data sources and embodiments, we are witnessing the emergence of agentic approaches that allow robots to reason their way through complex, unseen tasks. We propose to critically examine how these paradigms shift the traditional boundaries of robot learning:\n\n- Are we moving past narrow task-specific fine-tuning toward reasoning-based physical agents?\n- What are the implications of scaling laws, diverse training datasets, and multi-modal models in real-world deployment?\n- Can generalization alone lead us to the strong performance levels required in robotics use cases?\n\nWe seek diverse perspectives from the machine learning and robotics communities—both academia and industry—to map the trajectory of zero-shot physical AI.\n\nImportant Dates:\n- Submission Deadline: August 26, 2026 (11:59 PM AoE)\n- Decisions Announced: September 29, 2026\n\nSubmit via OpenReview." }, { "key": "Africa_in_AI", "title": "Africa in AI Affinity Group at Neural Information Processing Systems (NeurIPS) 2026", "subtitle": "Africa at NeurIPS 2026", "summary": "The 1st Africa in AI Workshop (\"Roots & Reach\") at NeurIPS 2026 elevates African perspectives on AI and sustainable development, welcoming research and position/vision papers across computer vision, LLMs, robotics, multi-agent systems, and ML that require at least one author working or studying in Africa.", "cfp_full": "1st Africa in AI Workshop - NeurIPS 2026\nROOTS & REACH - Scaling African AI for global impact.\nAFRICA IN AI - ROOTED LOCALLY, REACHING GLOBALLY\n\nWHEN: 6-12 December 2026\nWHERE: Sydney, Australia + Online\nFORMAT: Hybrid Workshop\n\n01 / THE WORKSHOP - Local knowledge. Global consequence.\n\nAfrica is not waiting for the future of AI. It is building it on its own terms, for its own realities, with lessons for the world.\n\nThe Africa in AI Affinity Group brings researchers, practitioners, industry partners and policymakers together at the intersection of artificial intelligence and sustainable development. At NeurIPS 2026, we will elevate African perspectives, unlock cross-disciplinary collaboration and advance trustworthy, resource-efficient AI.\n\n\"Roots\" grounds us in African contexts. \"Reach\" carries those ideas into the global AI ecosystem.\n\n02 / RESEARCH PILLARS - Research that moves communities forward.\n\nWe welcome original research and position or vision papers spanning computer vision, large language models, robotics, multi-agent systems and AI/ML.\n\n01. Healthy lives & a healthy planet - Biomedical and clinical AI, climate resilience, agriculture and tools that protect communities and ecosystems.\n02. Education & the future of work - Language technologies, cultural heritage, learning systems and pathways to productive, inclusive economies.\n03. Energy & thriving communities - Clean energy, safe water, resilient environments and intelligent infrastructure built for African contexts.\n\nWE ARE PARTICULARLY INTERESTED IN:\n- Low-resource, high-impact: Breakthroughs for African languages and locally relevant datasets.\n- Innovation infrastructure: Sustainable research labs and practical responses to compute constraints.\n- Non-WEIRD algorithms: Data sovereignty, AI ethics and models that move beyond Western defaults.\n- From lab to market: Routes to deploy and commercialise African AI in health, agriculture, language and fintech.\n\n03 / IMPORTANT DATES - Mark your calendar. All deadlines are 23:59 Anywhere on Earth unless otherwise stated.\n- 20 JUL 2026: Submission site opens\n- 10 AUG: Early deadline (Visa support track)\n- 10 SEP: Full paper deadline (11:59 PM AoE)\n- 20 SEP: Acceptance notification (All submissions)\n- 06 NOV 2026: Camera-ready deadline\n\n04 / CALL FOR PAPERS - Bring your work to the world.\n\nSubmissions must be original, include at least one researcher working or studying in Africa, and follow the published 2026 workshop format.\n\n01. Full paper - 8 pages of content plus up to 2 pages of references, submitted as a PDF.\n02. Double-blind review - Remove names, affiliations, websites, funding sources and identifying acknowledgements.\n03. Original work - Submissions must not have appeared in another venue, proceeding or journal.\n04. Submit on OpenReview - Upload your anonymised PDF through the official Africa in AI submission portal.\n\nTRAVEL AWARDS - Support to reach Sydney.\n\nA limited number of awards will help exceptional African researchers with conference registration, accommodation, flights and presentation development. Applicants must submit a workshop paper and be available for time-sensitive visa processing. Application deadline: 20 July 2026, 23:59 AoE.\n\nAfrica in AI Group and its events adhere to the NeurIPS Code of Conduct Policy and Procedures.", "cfp_status": "published", "topics": [ "Healthy lives & a healthy planet (biomedical and clinical AI, climate resilience, agriculture, tools that protect communities and ecosystems)", "Education & the future of work (language technologies, cultural heritage, learning systems, inclusive economies)", "Energy & thriving communities (clean energy, safe water, resilient environments, intelligent infrastructure for African contexts)", "Low-resource, high-impact work for African languages and locally relevant datasets", "Innovation infrastructure: sustainable research labs and responses to compute constraints", "Non-WEIRD algorithms: data sovereignty, AI ethics, models beyond Western defaults", "From lab to market: deploying and commercialising African AI in health, agriculture, language and fintech", "Computer vision, large language models, robotics, multi-agent systems, and AI/ML" ], "important_dates": [ { "label": "Submission site opens", "date": "2026-07-20" }, { "label": "Early deadline (Visa support track)", "date": "2026-08-10" }, { "label": "Full paper deadline", "date": "2026-09-10" }, { "label": "Acceptance notification", "date": "2026-09-20" }, { "label": "Camera-ready deadline", "date": "2026-11-06" }, { "label": "Travel award application deadline", "date": "2026-07-20" } ], "organizers": [ "Dr. Maruf Adewole (Medical Artificial Intelligence Laboratory, Nigeria)", "Dr. Udunna Anazodo (McGill University & MAI Lab, Nigeria)", "Dr. Sahar Selim (Center for Informatics Science, Giza, Egypt)", "Fatade Olawayemisi Boye (Babcock University, Nigeria)", "Lukman Enegi Ismaila (Johns Hopkins University School of Medicine, MD USA)", "Ifeoluwa Oladeji (Medical Artificial Intelligence Laboratory, Nigeria)" ], "speakers": [], "host_url": "https://africainai.mailab.io/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/Africa_in_AI", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/Africa_in_AI", "location": "Sydney, Australia", "city": "Sydney", "workshop_date": "", "contact": "info@mailab.io", "tracks": [ { "key": "Africa_in_AI", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/Africa_in_AI", "submission_dates_raw": "Submission Start: Jul 20 2026 12:00AM UTC-0, Submission Deadline: Aug 10 2026 12:00AM UTC-0" } ], "group": "society", "group_label": "Society, Applications & Community", "corpus": "Africa in AI Affinity Group at Neural Information Processing Systems (NeurIPS) 2026 Africa at NeurIPS 2026 The 1st Africa in AI Workshop (\"Roots & Reach\") at NeurIPS 2026 elevates African perspectives on AI and sustainable development, welcoming research and position/vision papers across computer vision, LLMs, robotics, multi-agent systems, and ML that require at least one author working or studying in Africa. Healthy lives & a healthy planet (biomedical and clinical AI, climate resilience, agriculture, tools that protect communities and ecosystems) Education & the future of work (language technologies, cultural heritage, learning systems, inclusive economies) Energy & thriving communities (clean energy, safe water, resilient environments, intelligent infrastructure for African contexts) Low-resource, high-impact work for African languages and locally relevant datasets Innovation infrastructure: sustainable research labs and responses to compute constraints Non-WEIRD algorithms: data sovereignty, AI ethics, models beyond Western defaults From lab to market: deploying and commercialising African AI in health, agriculture, language and fintech Computer vision, large language models, robotics, multi-agent systems, and AI/ML 1st Africa in AI Workshop - NeurIPS 2026\nROOTS & REACH - Scaling African AI for global impact.\nAFRICA IN AI - ROOTED LOCALLY, REACHING GLOBALLY\n\nWHEN: 6-12 December 2026\nWHERE: Sydney, Australia + Online\nFORMAT: Hybrid Workshop\n\n01 / THE WORKSHOP - Local knowledge. Global consequence.\n\nAfrica is not waiting for the future of AI. It is building it on its own terms, for its own realities, with lessons for the world.\n\nThe Africa in AI Affinity Group brings researchers, practitioners, industry partners and policymakers together at the intersection of artificial intelligence and sustainable development. At NeurIPS 2026, we will elevate African perspectives, unlock cross-disciplinary collaboration and advance trustworthy, resource-efficient AI.\n\n\"Roots\" grounds us in African contexts. \"Reach\" carries those ideas into the global AI ecosystem.\n\n02 / RESEARCH PILLARS - Research that moves communities forward.\n\nWe welcome original research and position or vision papers spanning computer vision, large language models, robotics, multi-agent systems and AI/ML.\n\n01. Healthy lives & a healthy planet - Biomedical and clinical AI, climate resilience, agriculture and tools that protect communities and ecosystems.\n02. Education & the future of work - Language technologies, cultural heritage, learning systems and pathways to productive, inclusive economies.\n03. Energy & thriving communities - Clean energy, safe water, resilient environments and intelligent infrastructure built for African contexts.\n\nWE ARE PARTICULARLY INTERESTED IN:\n- Low-resource, high-impact: Breakthroughs for African languages and locally relevant datasets.\n- Innovation infrastructure: Sustainable research labs and practical responses to compute constraints.\n- Non-WEIRD algorithms: Data sovereignty, AI ethics and models that move beyond Western defaults.\n- From lab to market: Routes to deploy and commercialise African AI in health, agriculture, language and fintech.\n\n03 / IMPORTANT DATES - Mark your calendar. All deadlines are 23:59 Anywhere on Earth unless otherwise stated.\n- 20 JUL 2026: Submission site opens\n- 10 AUG: Early deadline (Visa support track)\n- 10 SEP: Full paper deadline (11:59 PM AoE)\n- 20 SEP: Acceptance notification (All submissions)\n- 06 NOV 2026: Camera-ready deadline\n\n04 / CALL FOR PAPERS - Bring your work to the world.\n\nSubmissions must be original, include at least one researcher working or studying in Africa, and follow the published 2026 workshop format.\n\n01. Full paper - 8 pages of content plus up to 2 pages of references, submitted as a PDF.\n02. Double-blind review - Remove names, affiliations, websites, funding sources and identifying acknowledgements.\n03. Original work - Submissions must not have appeared in another venue, proceeding or journal.\n04. Submit on OpenReview - Upload your anonymised PDF through the official Africa in AI submission portal.\n\nTRAVEL AWARDS - Support to reach Sydney.\n\nA limited number of awards will help exceptional African researchers with conference registration, accommodation, flights and presentation development. Applicants must submit a workshop paper and be available for time-sensitive visa processing. Application deadline: 20 July 2026, 23:59 AoE.\n\nAfrica in AI Group and its events adhere to the NeurIPS Code of Conduct Policy and Procedures." }, { "key": "AISciK", "title": "AI & Science: Evolution or Extinction?", "subtitle": "AISciK 2026", "summary": "An interdisciplinary workshop toward a domain-specific safety, alignment, and evaluation agenda for AI in science, gathering researchers from statistics, philosophy and sociology of science, science and technology studies, psychology, and anthropology alongside AI researchers to define scientific integrity, evaluations, and sociotechnical guardrails for human-AI scientific collaboration.", "cfp_full": "About the workshop\n\nRecent years have shown an explosion of interest for supporting, accelerating, and automating scientific discovery via AI systems. Both academic and industry AI researchers have leapt to the wellspring of challenging computational problems currently unsolved by the scientific community as a way to test the state of the art. This broad appeal has translated to a plethora of interdisciplinary collaborations between AI researchers and traditional scientists across multiple fields of science, all aiming to highlight the potential of AI models to contribute to scientific research.\n\nIn order to properly evaluate how AI systems can impact the practice of science, we first must agree upon consistent definitions of success that outline how this type of integration can occur safely. Our workshop will gather researchers from statistics, philosophy of science, sociology of science, science and technology studies, psychology, and anthropology, alongside AI researchers working on AI for science, interpretability, AI safety, and agent evaluations, to address three questions:\n\n- What epistemic values constitute scientific integrity in the era of human-AI collaboration?\n- How can we build evaluations that measure whether AI systems uphold these values in practice?\n- What sociotechnical guardrails can sustain robust human-AI scientific collaboration without eroding the integrity of scientific knowledge production?\n\nAnswering any of these requires expertise that no single community currently holds. Our workshop aims to both initiate a much-needed conversation for the future of the scientific and AI communities as well as foster a community of like-minded researchers committed to addressing these questions long-term in an interdisciplinary fashion.\n\nCall for papers\n\nTo Be Announced\n\nSubmission guidelines\n\nTo Be Announced\n\nImportant dates\n\n- Submission deadline: August 29, 2026\n- Notification of acceptance: September 29, 2026\n- Camera-ready deadline: TBA\n- Workshop date: TBA\n\nAll deadlines are 23:59, Anywhere on Earth (AoE).", "cfp_status": "published", "topics": [ "Epistemic values and scientific integrity in the era of human-AI collaboration", "Evaluations measuring whether AI systems uphold scientific values in practice", "Sociotechnical guardrails for robust human-AI scientific collaboration", "AI for science, interpretability, AI safety, and agent evaluations", "Philosophy, sociology, and science and technology studies of AI in science" ], "important_dates": [ { "label": "Submission deadline", "date": "2026-08-29" }, { "label": "Notification of acceptance", "date": "2026-09-29" }, { "label": "Camera-ready deadline", "date": "TBA" }, { "label": "Workshop date", "date": "TBA" } ], "organizers": [ "Savannah Thais (City University of New York, USA)", "Nathan Suri (Yale University, USA)", "Lauren Greenspan (Principles of Intelligence (PrincInt), USA)", "Max Hennick (Stormglass AI, Canada)", "Roberto Trotta (International School for Advanced Studies (SISSA), Italy)" ], "speakers": [ "Ranjit Singh (Data & Society)", "Federico Bianchi (Together AI)", "Yian Yin (Cornell)", "Mel Andrews (Princeton)", "Daniel Herrmann (UNC)", "Lisa Messeri (Yale)", "David Hogg (NYU)", "Jesse Thaler (MIT)" ], "host_url": "https://aiscik.github.io/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AISciK", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AISciK", "location": "Atlanta, Georgia, USA", "city": "Atlanta", "workshop_date": "2026-12-12", "contact": "aiscik.workshop@gmail.com", "tracks": [ { "key": "AISciK", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AISciK", "submission_dates_raw": "" } ], "group": "science", "group_label": "AI for Science", "corpus": "AI & Science: Evolution or Extinction? AISciK 2026 An interdisciplinary workshop toward a domain-specific safety, alignment, and evaluation agenda for AI in science, gathering researchers from statistics, philosophy and sociology of science, science and technology studies, psychology, and anthropology alongside AI researchers to define scientific integrity, evaluations, and sociotechnical guardrails for human-AI scientific collaboration. Epistemic values and scientific integrity in the era of human-AI collaboration Evaluations measuring whether AI systems uphold scientific values in practice Sociotechnical guardrails for robust human-AI scientific collaboration AI for science, interpretability, AI safety, and agent evaluations Philosophy, sociology, and science and technology studies of AI in science About the workshop\n\nRecent years have shown an explosion of interest for supporting, accelerating, and automating scientific discovery via AI systems. Both academic and industry AI researchers have leapt to the wellspring of challenging computational problems currently unsolved by the scientific community as a way to test the state of the art. This broad appeal has translated to a plethora of interdisciplinary collaborations between AI researchers and traditional scientists across multiple fields of science, all aiming to highlight the potential of AI models to contribute to scientific research.\n\nIn order to properly evaluate how AI systems can impact the practice of science, we first must agree upon consistent definitions of success that outline how this type of integration can occur safely. Our workshop will gather researchers from statistics, philosophy of science, sociology of science, science and technology studies, psychology, and anthropology, alongside AI researchers working on AI for science, interpretability, AI safety, and agent evaluations, to address three questions:\n\n- What epistemic values constitute scientific integrity in the era of human-AI collaboration?\n- How can we build evaluations that measure whether AI systems uphold these values in practice?\n- What sociotechnical guardrails can sustain robust human-AI scientific collaboration without eroding the integrity of scientific knowledge production?\n\nAnswering any of these requires expertise that no single community currently holds. Our workshop aims to both initiate a much-needed conversation for the future of the scientific and AI communities as well as foster a community of like-minded researchers committed to addressing these questions long-term in an interdisciplinary fashion.\n\nCall for papers\n\nTo Be Announced\n\nSubmission guidelines\n\nTo Be Announced\n\nImportant dates\n\n- Submission deadline: August 29, 2026\n- Notification of acceptance: September 29, 2026\n- Camera-ready deadline: TBA\n- Workshop date: TBA\n\nAll deadlines are 23:59, Anywhere on Earth (AoE)." }, { "key": "AIM", "title": "AI Agents for Biomedical Imaging and Multimodal Clinical Data at NeurIPS2026", "subtitle": "AIM NeurIPS2026", "summary": "A one-day workshop on AI agents for biomedical imaging and multimodal clinical data, focused on how learning systems can organize multi-step biomedical analyses around images, clinical context, specialized tools, intermediate evidence, and human feedback, and on defining shared tasks, evaluation protocols, and benchmarks for agentic image-analysis systems.", "cfp_full": "About the Workshop\n\nBiomedical imaging AI has produced strong methods for segmentation, registration, reconstruction, detection, classification, and report generation. Yet most systems remain organized around fixed inputs and outputs - a structure that fails to capture real clinical workflows, where experts combine images with reports, prior studies, measurements, laboratory values, EHR data, waveforms, pathology, video, and longitudinal patient context.\n\nThis workshop focuses on AI agents for biomedical imaging and multimodal clinical data. The central theme is how learning systems can organize multi-step biomedical analyses around images, clinical context, specialized tools, intermediate evidence, and human feedback - for example, systems that retrieve prior studies, invoke segmentation or registration software, measure structures or lesions, integrate reports or EHR variables, compare data across time, and expose intermediate outputs for review.\n\nThe workshop brings together researchers in machine learning, computer vision, biomedical imaging, multimodal learning, clinical AI, medical image computing, biomedical informatics, and trustworthy AI to define shared task definitions, evaluation protocols, and benchmarks for agentic image-analysis systems. It is organized around a set of concrete open questions:\n\n- How should agents invoke and sequence specialized image-analysis tools - segmentation, registration, reconstruction - in ways that are reliable, auditable, and clinically meaningful?\n- How should rich multimodal clinical context (reports, EHR variables, laboratory findings, waveforms, longitudinal records) be grounded in image evidence?\n- How should intermediate outputs of multi-step analyses be surfaced, verified, and corrected before downstream steps consume them?\n- How should full systems be evaluated - not just on final accuracy, but on reasoning quality, uncertainty calibration, error recovery, robustness, and readiness for human oversight?\n\nTopics of Interest\n\n- AI agents for image analysis, measurement, annotation, reporting, and cohort discovery\n- Multimodal systems combining images with reports, EHR data, notes, waveforms, omics, video, or longitudinal records\n- Tool-using systems that invoke segmentation, registration, reconstruction, retrieval, visualization, uncertainty estimation, or statistical analysis\n- Evaluation of grounding, reasoning, uncertainty, error recovery, robustness, and reproducibility\n- Human-in-the-loop systems for expert guidance and quality control\n- Datasets, benchmarks, and software, and position papers defining reusable tasks for the community\n\nCall for Papers\n\nNon-archival submissions through OpenReview, with an optional archival pathway via the MELBA special issue.\n\nWe solicit submissions on all topics related to AI agents for biomedical imaging and multimodal clinical data (see topics above). Appropriate submissions include methods papers, benchmark or dataset papers, software/resource papers, demos, and position papers.\n\nSubmission Tracks\n\n- Full papers - up to 9 pages (tentative; final page limits TBA), with unlimited references.\n- Short papers - up to 4 pages (tentative; final page limits TBA), with unlimited references.\n\nPapers must use the NeurIPS 2026 paper template and guidelines.\n\nReview and Presentation\n\n- Submissions are made through the workshop's OpenReview venue; each submission will receive three reviews whenever possible.\n- Review criteria: relevance to agentic biomedical imaging or multimodal clinical data, technical quality, clarity, evaluation rigor, reproducibility, and appropriateness for workshop discussion.\n- Accepted submissions will be presented as posters, demos, or contributed talks.\n- Accepted papers will be hosted on OpenReview and linked and hosted directly on this workshop website.\n- The workshop is non-archival.\n- Organizers follow NeurIPS conflict-of-interest rules and will not assess submissions from conflicted authors.\n\nMELBA Special Issue\n\nThe workshop is paired with a special issue of the Machine Learning for Biomedical Imaging (MELBA) journal on AI agents for biomedical imaging and multimodal clinical data. Authors of accepted workshop papers will be invited to submit extended journal-length manuscripts to the special issue. Submission to MELBA is optional and follows MELBA's independent editorial and peer-review process, providing an archival pathway while preserving the non-archival status of workshop submissions.\n\nImportant Dates\n\n- Call for papers released: July 2026\n- Paper submission deadline: August 29, 2026 (AoE)\n- Notification of acceptance: No later than September 29, 2026\n- Public schedule posted: October 2026\n- NeurIPS workshop: December 12 or 13, 2026 (final date TBA)\n- MELBA special issue submissions: Target January 2027\n\nAll deadlines are Anywhere on Earth (AoE, UTC-12).\n\nChecklist\n\n- Use the NeurIPS 2026 LaTeX template and guidelines.\n- Anonymize your submission (double-blind review).\n- Full papers: up to 9 pages; short papers: up to 4 pages (tentative; final limits TBA - unlimited references).\n- Accepted papers will be hosted on OpenReview and on this website; submissions are non-archival, and extended versions may later be submitted to the MELBA special issue.\n- Submit by August 29, 2026 (AoE).\n\nNote on OpenReview profiles: new profiles created with an institutional email are activated automatically; profiles created without an institutional email go through a moderation process that can take up to two weeks. Please create your OpenReview profile well before the deadline.", "cfp_status": "published", "topics": [ "AI agents for image analysis, measurement, annotation, reporting, and cohort discovery", "Multimodal systems combining images with reports, EHR data, notes, waveforms, omics, video, or longitudinal records", "Tool-using systems that invoke segmentation, registration, reconstruction, retrieval, visualization, uncertainty estimation, or statistical analysis", "Evaluation of grounding, reasoning, uncertainty, error recovery, robustness, and reproducibility", "Human-in-the-loop systems for expert guidance and quality control", "Datasets, benchmarks, and software, and position papers defining reusable tasks for the community" ], "important_dates": [ { "label": "Call for papers released", "date": "July 2026" }, { "label": "Paper submission deadline", "date": "2026-08-29" }, { "label": "Notification of acceptance", "date": "2026-09-29" }, { "label": "Public schedule posted", "date": "October 2026" }, { "label": "NeurIPS workshop", "date": "2026-12-12" }, { "label": "MELBA special issue submissions", "date": "January 2027" } ], "organizers": [ "Ehsan Adeli (Stanford University)", "Tal Arbel (McGill University)", "Adrian V. Dalca (MGH / Harvard Medical School / MIT)", "Klaus Maier-Hein (German Cancer Research Center (DKFZ))", "Yixuan Yuan (The Chinese University of Hong Kong)" ], "speakers": [], "host_url": "https://aim-neurips26.github.io/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AIM", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AIM", "location": "Atlanta, Georgia", "city": "Atlanta", "workshop_date": "2026-12-12", "contact": "eadeli@stanford.edu", "tracks": [ { "key": "AIM", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AIM", "submission_dates_raw": "" } ], "group": "health", "group_label": "Health & Medicine", "corpus": "AI Agents for Biomedical Imaging and Multimodal Clinical Data at NeurIPS2026 AIM NeurIPS2026 A one-day workshop on AI agents for biomedical imaging and multimodal clinical data, focused on how learning systems can organize multi-step biomedical analyses around images, clinical context, specialized tools, intermediate evidence, and human feedback, and on defining shared tasks, evaluation protocols, and benchmarks for agentic image-analysis systems. AI agents for image analysis, measurement, annotation, reporting, and cohort discovery Multimodal systems combining images with reports, EHR data, notes, waveforms, omics, video, or longitudinal records Tool-using systems that invoke segmentation, registration, reconstruction, retrieval, visualization, uncertainty estimation, or statistical analysis Evaluation of grounding, reasoning, uncertainty, error recovery, robustness, and reproducibility Human-in-the-loop systems for expert guidance and quality control Datasets, benchmarks, and software, and position papers defining reusable tasks for the community About the Workshop\n\nBiomedical imaging AI has produced strong methods for segmentation, registration, reconstruction, detection, classification, and report generation. Yet most systems remain organized around fixed inputs and outputs - a structure that fails to capture real clinical workflows, where experts combine images with reports, prior studies, measurements, laboratory values, EHR data, waveforms, pathology, video, and longitudinal patient context.\n\nThis workshop focuses on AI agents for biomedical imaging and multimodal clinical data. The central theme is how learning systems can organize multi-step biomedical analyses around images, clinical context, specialized tools, intermediate evidence, and human feedback - for example, systems that retrieve prior studies, invoke segmentation or registration software, measure structures or lesions, integrate reports or EHR variables, compare data across time, and expose intermediate outputs for review.\n\nThe workshop brings together researchers in machine learning, computer vision, biomedical imaging, multimodal learning, clinical AI, medical image computing, biomedical informatics, and trustworthy AI to define shared task definitions, evaluation protocols, and benchmarks for agentic image-analysis systems. It is organized around a set of concrete open questions:\n\n- How should agents invoke and sequence specialized image-analysis tools - segmentation, registration, reconstruction - in ways that are reliable, auditable, and clinically meaningful?\n- How should rich multimodal clinical context (reports, EHR variables, laboratory findings, waveforms, longitudinal records) be grounded in image evidence?\n- How should intermediate outputs of multi-step analyses be surfaced, verified, and corrected before downstream steps consume them?\n- How should full systems be evaluated - not just on final accuracy, but on reasoning quality, uncertainty calibration, error recovery, robustness, and readiness for human oversight?\n\nTopics of Interest\n\n- AI agents for image analysis, measurement, annotation, reporting, and cohort discovery\n- Multimodal systems combining images with reports, EHR data, notes, waveforms, omics, video, or longitudinal records\n- Tool-using systems that invoke segmentation, registration, reconstruction, retrieval, visualization, uncertainty estimation, or statistical analysis\n- Evaluation of grounding, reasoning, uncertainty, error recovery, robustness, and reproducibility\n- Human-in-the-loop systems for expert guidance and quality control\n- Datasets, benchmarks, and software, and position papers defining reusable tasks for the community\n\nCall for Papers\n\nNon-archival submissions through OpenReview, with an optional archival pathway via the MELBA special issue.\n\nWe solicit submissions on all topics related to AI agents for biomedical imaging and multimodal clinical data (see topics above). Appropriate submissions include methods papers, benchmark or dataset papers, software/resource papers, demos, and position papers.\n\nSubmission Tracks\n\n- Full papers - up to 9 pages (tentative; final page limits TBA), with unlimited references.\n- Short papers - up to 4 pages (tentative; final page limits TBA), with unlimited references.\n\nPapers must use the NeurIPS 2026 paper template and guidelines.\n\nReview and Presentation\n\n- Submissions are made through the workshop's OpenReview venue; each submission will receive three reviews whenever possible.\n- Review criteria: relevance to agentic biomedical imaging or multimodal clinical data, technical quality, clarity, evaluation rigor, reproducibility, and appropriateness for workshop discussion.\n- Accepted submissions will be presented as posters, demos, or contributed talks.\n- Accepted papers will be hosted on OpenReview and linked and hosted directly on this workshop website.\n- The workshop is non-archival.\n- Organizers follow NeurIPS conflict-of-interest rules and will not assess submissions from conflicted authors.\n\nMELBA Special Issue\n\nThe workshop is paired with a special issue of the Machine Learning for Biomedical Imaging (MELBA) journal on AI agents for biomedical imaging and multimodal clinical data. Authors of accepted workshop papers will be invited to submit extended journal-length manuscripts to the special issue. Submission to MELBA is optional and follows MELBA's independent editorial and peer-review process, providing an archival pathway while preserving the non-archival status of workshop submissions.\n\nImportant Dates\n\n- Call for papers released: July 2026\n- Paper submission deadline: August 29, 2026 (AoE)\n- Notification of acceptance: No later than September 29, 2026\n- Public schedule posted: October 2026\n- NeurIPS workshop: December 12 or 13, 2026 (final date TBA)\n- MELBA special issue submissions: Target January 2027\n\nAll deadlines are Anywhere on Earth (AoE, UTC-12).\n\nChecklist\n\n- Use the NeurIPS 2026 LaTeX template and guidelines.\n- Anonymize your submission (double-blind review).\n- Full papers: up to 9 pages; short papers: up to 4 pages (tentative; final limits TBA - unlimited references).\n- Accepted papers will be hosted on OpenReview and on this website; submissions are non-archival, and extended versions may later be submitted to the MELBA special issue.\n- Submit by August 29, 2026 (AoE).\n\nNote on OpenReview profiles: new profiles created with an institutional email are activated automatically; profiles created without an institutional email go through a moderation process that can take up to two weeks. Please create your OpenReview profile well before the deadline." }, { "key": "AI_and_the_Self", "title": "AI and the Self: Human Identity, Authenticity, and Agency in the Age of AI", "subtitle": "AI and the Self 2026", "summary": "A NeurIPS 2026 workshop bringing together technical, empirical, social, cultural, and philosophical work on how AI systems mediate identity, agency, self-understanding, dependence, and human flourishing.", "cfp_full": "AI and the Self — Human Identity, Authenticity, and Agency in the Age of AI\n\nAI and the Self brings together technical, empirical, social, cultural, and philosophical work on how AI systems mediate identity, agency, self-understanding, dependence, and human flourishing.\n\nAbout / Workshop overview\n\nAI systems are no longer only external tools. They are increasingly becoming interlocutors, cognitive extensions, mirrors, collaborators, and social actors that shape how people think, decide, remember, and understand themselves.\n\nThis workshop asks how AI systems construct, extend, support, or destabilize the self, and what kinds of computational, empirical, and interdisciplinary methods are needed to study those effects rigorously.\n\nOne central aim is to move these questions closer to machine learning practice: how to measure shifts in self-concept, dependence, confidence, preference formation, and memory under repeated AI interaction; how to design personalization, memory, uncertainty, and refusal in ways that support reflection rather than over-reliance; and how to study cultural and linguistic assumptions about the self inside AI systems.\n\nThe workshop also looks at how AI changes learning, authorship, creativity, professional identity, and the meaning of human competence when systems act as tutors, collaborators, co-authors, or social companions. It builds on the earlier interdisciplinary workshop in Bonn and translates that conversation into a NeurIPS-facing research agenda.\n\nResearch themes\n- Measurement and evaluation\n- Design and interaction\n- Culture and language\n- Education and authorship\n- Ethics and philosophy\n\nThe workshop welcomes technical, empirical, socio-cultural, philosophical, and application-oriented work, including work in progress and new problem framings.\n\nCall for Papers\n\nThe workshop seeks submissions at the intersection of AI/ML and human selfhood, spanning technical, empirical, and humanistic perspectives. Research interests include identity, agency, authenticity, dependence, self-modeling, memory, authorship, value change, culture, and AI-mediated reflection.\n\nSubmission types:\n- Long papers: up to 8 pages of content with unlimited references and appendix\n- Short papers: up to 4 pages of content with unlimited references and appendix\n\nAll submissions must be anonymized for double-blind review and are non-archival. Work in progress, early findings, and interdisciplinary themes are welcome. Submissions go through the OpenReview submission portal.\n\nKey Dates (all deadlines use 23:59 Anywhere on Earth (AoE)):\n- Submissions open: July 24, 2026\n- Submission deadline: August 24, 2026\n- Acceptance notifications: September 28, 2026\n- Workshop date: December 12 or 13, 2026 (specific day pending confirmation), Paris", "cfp_status": "published", "topics": [ "Measurement and evaluation of shifts in self-concept, dependence, confidence, preference formation, and memory under repeated AI interaction", "Design and interaction (personalization, memory, uncertainty, and refusal that support reflection rather than over-reliance)", "Culture and language (cultural and linguistic assumptions about the self inside AI systems)", "Education and authorship (learning, authorship, creativity, professional identity, human competence)", "Ethics and philosophy", "Identity, agency, authenticity, dependence, self-modeling, memory, value change, and AI-mediated reflection" ], "important_dates": [ { "label": "Submissions open", "date": "2026-07-24" }, { "label": "Submission deadline", "date": "2026-08-24" }, { "label": "Acceptance notifications", "date": "2026-09-28" }, { "label": "Workshop", "date": "2026-12-12" } ], "organizers": [], "speakers": [], "host_url": "https://aintheself.github.io/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AI_and_the_Self", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AI_and_the_Self", "location": "Paris, France", "city": "Paris", "workshop_date": "2026-12-11", "contact": "aintheself@gmail.com", "tracks": [ { "key": "AI_and_the_Self", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AI_and_the_Self", "submission_dates_raw": "" } ], "group": "society", "group_label": "Society, Applications & Community", "corpus": "AI and the Self: Human Identity, Authenticity, and Agency in the Age of AI AI and the Self 2026 A NeurIPS 2026 workshop bringing together technical, empirical, social, cultural, and philosophical work on how AI systems mediate identity, agency, self-understanding, dependence, and human flourishing. Measurement and evaluation of shifts in self-concept, dependence, confidence, preference formation, and memory under repeated AI interaction Design and interaction (personalization, memory, uncertainty, and refusal that support reflection rather than over-reliance) Culture and language (cultural and linguistic assumptions about the self inside AI systems) Education and authorship (learning, authorship, creativity, professional identity, human competence) Ethics and philosophy Identity, agency, authenticity, dependence, self-modeling, memory, value change, and AI-mediated reflection AI and the Self — Human Identity, Authenticity, and Agency in the Age of AI\n\nAI and the Self brings together technical, empirical, social, cultural, and philosophical work on how AI systems mediate identity, agency, self-understanding, dependence, and human flourishing.\n\nAbout / Workshop overview\n\nAI systems are no longer only external tools. They are increasingly becoming interlocutors, cognitive extensions, mirrors, collaborators, and social actors that shape how people think, decide, remember, and understand themselves.\n\nThis workshop asks how AI systems construct, extend, support, or destabilize the self, and what kinds of computational, empirical, and interdisciplinary methods are needed to study those effects rigorously.\n\nOne central aim is to move these questions closer to machine learning practice: how to measure shifts in self-concept, dependence, confidence, preference formation, and memory under repeated AI interaction; how to design personalization, memory, uncertainty, and refusal in ways that support reflection rather than over-reliance; and how to study cultural and linguistic assumptions about the self inside AI systems.\n\nThe workshop also looks at how AI changes learning, authorship, creativity, professional identity, and the meaning of human competence when systems act as tutors, collaborators, co-authors, or social companions. It builds on the earlier interdisciplinary workshop in Bonn and translates that conversation into a NeurIPS-facing research agenda.\n\nResearch themes\n- Measurement and evaluation\n- Design and interaction\n- Culture and language\n- Education and authorship\n- Ethics and philosophy\n\nThe workshop welcomes technical, empirical, socio-cultural, philosophical, and application-oriented work, including work in progress and new problem framings.\n\nCall for Papers\n\nThe workshop seeks submissions at the intersection of AI/ML and human selfhood, spanning technical, empirical, and humanistic perspectives. Research interests include identity, agency, authenticity, dependence, self-modeling, memory, authorship, value change, culture, and AI-mediated reflection.\n\nSubmission types:\n- Long papers: up to 8 pages of content with unlimited references and appendix\n- Short papers: up to 4 pages of content with unlimited references and appendix\n\nAll submissions must be anonymized for double-blind review and are non-archival. Work in progress, early findings, and interdisciplinary themes are welcome. Submissions go through the OpenReview submission portal.\n\nKey Dates (all deadlines use 23:59 Anywhere on Earth (AoE)):\n- Submissions open: July 24, 2026\n- Submission deadline: August 24, 2026\n- Acceptance notifications: September 28, 2026\n- Workshop date: December 12 or 13, 2026 (specific day pending confirmation), Paris" }, { "key": "ASCI", "title": "AI at Scale for Clinical Impact: Cancer Pathology Foundation Models", "subtitle": "ASCI", "summary": "A NeurIPS 2026 workshop bringing together researchers, clinicians, and industry teams building cancer AI models that move from technical benchmarks to validated clinical impact, with emphasis on hospital-scale pathology foundation models, multimodal clinical data, evaluation, and translation into practice.", "cfp_full": "AI at Scale for Clinical Impact (ASCI): Cancer Pathology Foundation Models\n\nASCI is a NeurIPS 2026 workshop in Sydney, Australia for researchers, clinicians, and industry teams building cancer AI models that can move from technical benchmarks to validated clinical impact.\n\nCall for Submitted Manuscripts\n\nClinical-scale pathology foundation models\n\nASCI invites submissions on cancer AI models and beyond, with emphasis on hospital-scale learning, multimodal clinical data, evaluation, and translation into practice.\n\n- Pathology-aware learning\n- Continual learning at hospital scale\n- Few-shot adaptation for rare cancers and emerging biomarkers\n- Multimodal integration across histology, omics, spatial data, and text\n- Evaluation, benchmarking, reporting standards, and clinical validation\n- Interpretability, biological insight, and real-world deployment\n\nSubmission Instructions\n\nSubmit on OpenReview ASCI workshop page. Please see the NeurIPS website for more information.\n\nAbstract registration (required): August 20, 2026, 12:00 AM UTC.\n\nFull paper deadline: August 29, 2026, 12:00 AM UTC.\n\nSubmit through the ASCI OpenReview workshop page.\n\nUse the NeurIPS 2026 paper formatting and LaTeX style files: download the official style package or open the Overleaf template.\n\nSubmit PDF manuscripts for single-blind review.\n\nThe main text of a submitted paper is limited to nine content pages, including all figures and tables. Additional pages containing references and the optional technical appendices do not count as content pages.\n\nReview and Presentation\n\nSubmissions will be reviewed for technical quality, clinical relevance, and fit with the workshop themes.\n\nAccepted manuscripts will be invited for oral spotlight or poster presentation.\n\nDates:\n- Abstract Registration (Required): August 20, 2026, 12:00 AM UTC\n- Full Paper Deadline: August 29, 2026, 12:00 AM UTC\n- Accept / Reject Decision: September 29, 2026\n\nThe workshop runs a full day from 8:00 AM to 5:00 PM at the International Convention Centre Sydney (ICC Sydney), 14 Darling Drive, Sydney NSW 2000, Australia.", "cfp_status": "published", "topics": [ "Pathology-aware learning", "Continual learning at hospital scale", "Few-shot adaptation for rare cancers and emerging biomarkers", "Multimodal integration across histology, omics, spatial data, and text", "Evaluation, benchmarking, reporting standards, and clinical validation", "Interpretability, biological insight, and real-world deployment" ], "important_dates": [ { "label": "Abstract Registration (Required)", "date": "2026-08-20" }, { "label": "Full Paper Deadline", "date": "2026-08-29" }, { "label": "Accept / Reject Decision", "date": "2026-09-29" }, { "label": "Workshop", "date": "2026-12-11" } ], "organizers": [ "Chad Vanderbilt, MD (Memorial Sloan Kettering Cancer Center) — Chair", "Neeraj Kumar, PhD (Memorial Sloan Kettering Cancer Center) — Chair", "Gabriele Campanella, PhD (Icahn School of Medicine at Mount Sinai)", "Ruchika Verma, PhD (Icahn School of Medicine at Mount Sinai)", "Jia Wu, PhD (MD Anderson Cancer Center)", "Jianjun Zhang, PhD (MD Anderson Cancer Center)", "Luisa M. Solis Soto, MD (MD Anderson Cancer Center)", "Rukhmini Bandyopadhyay (MD Anderson Cancer Center)", "Muhammad Waqas (MD Anderson Cancer Center)", "Hamid Reza Tizhoosh, PhD (Mayo Clinic)", "Wataru Uegami, MD, PhD (Mayo Clinic)", "Saghir A. Al-Fasly, PhD (Mayo Clinic)", "Joel Saltz, MD, PhD (Stony Brook University)", "Jakub Kaczmarzyk (Stony Brook University)", "Kostas Triaridis (Stony Brook University)", "Katherine Hoadley, PhD (University of North Carolina at Chapel Hill)", "Melissa Troester, PhD (University of North Carolina at Chapel Hill)", "Siddharth Singi, MS (Memorial Sloan Kettering Cancer Center)" ], "speakers": [ "Thomas Fuchs (Chief AI Officer, Eli Lilly, USA)", "Hoifung Poon (General Manager, Microsoft Research, USA)", "Piotr Bojanowski (Research Director, Meta, USA)", "Jana Lipkova (Assistant Professor, University of California - Irvine, USA)", "Ruijiang Li (Professor, Stanford University, USA)", "Jakob Nikolas Kather (Professor, Technical University Dresden, Germany)", "Bo Wang (Professor, University of Toronto, Canada)", "Jorge Reis-Filho (Chief of AI for Science Innovation, AstraZeneca, UK)", "Nigam Shah (Professor, Stanford University, USA)", "Danielle Bitterman (Vice President, AI for Clinical Development, AstraZeneca, UK)" ], "host_url": "https://asci.artificialintelligencepathology.org/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/ASCI", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/ASCI", "location": "Sydney, Australia", "city": "Sydney", "workshop_date": "", "contact": "kumarn6@mskcc.org", "tracks": [ { "key": "ASCI", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/ASCI", "submission_dates_raw": "Abstract Registration: Aug 20 2026 12:00AM UTC-0, Submission Deadline: Aug 29 2026 12:00AM UTC-0" } ], "group": "health", "group_label": "Health & Medicine", "corpus": "AI at Scale for Clinical Impact: Cancer Pathology Foundation Models ASCI A NeurIPS 2026 workshop bringing together researchers, clinicians, and industry teams building cancer AI models that move from technical benchmarks to validated clinical impact, with emphasis on hospital-scale pathology foundation models, multimodal clinical data, evaluation, and translation into practice. Pathology-aware learning Continual learning at hospital scale Few-shot adaptation for rare cancers and emerging biomarkers Multimodal integration across histology, omics, spatial data, and text Evaluation, benchmarking, reporting standards, and clinical validation Interpretability, biological insight, and real-world deployment AI at Scale for Clinical Impact (ASCI): Cancer Pathology Foundation Models\n\nASCI is a NeurIPS 2026 workshop in Sydney, Australia for researchers, clinicians, and industry teams building cancer AI models that can move from technical benchmarks to validated clinical impact.\n\nCall for Submitted Manuscripts\n\nClinical-scale pathology foundation models\n\nASCI invites submissions on cancer AI models and beyond, with emphasis on hospital-scale learning, multimodal clinical data, evaluation, and translation into practice.\n\n- Pathology-aware learning\n- Continual learning at hospital scale\n- Few-shot adaptation for rare cancers and emerging biomarkers\n- Multimodal integration across histology, omics, spatial data, and text\n- Evaluation, benchmarking, reporting standards, and clinical validation\n- Interpretability, biological insight, and real-world deployment\n\nSubmission Instructions\n\nSubmit on OpenReview ASCI workshop page. Please see the NeurIPS website for more information.\n\nAbstract registration (required): August 20, 2026, 12:00 AM UTC.\n\nFull paper deadline: August 29, 2026, 12:00 AM UTC.\n\nSubmit through the ASCI OpenReview workshop page.\n\nUse the NeurIPS 2026 paper formatting and LaTeX style files: download the official style package or open the Overleaf template.\n\nSubmit PDF manuscripts for single-blind review.\n\nThe main text of a submitted paper is limited to nine content pages, including all figures and tables. Additional pages containing references and the optional technical appendices do not count as content pages.\n\nReview and Presentation\n\nSubmissions will be reviewed for technical quality, clinical relevance, and fit with the workshop themes.\n\nAccepted manuscripts will be invited for oral spotlight or poster presentation.\n\nDates:\n- Abstract Registration (Required): August 20, 2026, 12:00 AM UTC\n- Full Paper Deadline: August 29, 2026, 12:00 AM UTC\n- Accept / Reject Decision: September 29, 2026\n\nThe workshop runs a full day from 8:00 AM to 5:00 PM at the International Convention Centre Sydney (ICC Sydney), 14 Darling Drive, Sydney NSW 2000, Australia." }, { "key": "AI4Mat", "title": "AI for Accelerated Materials Design - NeurIPS 2026", "subtitle": "AI4Mat-NeurIPS-2026", "summary": "An interdisciplinary workshop bringing AI researchers and materials scientists together to tackle challenges in AI-driven materials discovery and development, spanning AI-guided design, synthesis, and automated characterization.", "cfp_full": "AI4Mat: AI for Accelerated Materials Design\nDecember 2026 @ NeurIPS 2026 (Sydney, Australia)\n\nAbout the Workshop\nThe AI for Accelerated Materials Discovery (AI4Mat) Workshop at NeurIPS 2026 provides an inclusive and collaborative platform where AI researchers and material scientists converge to tackle the cutting-edge challenges in AI-driven materials discovery and development. Our goal is to foster a vibrant exchange of ideas, breaking down barriers between disciplines and encouraging insightful discussions among experts from diverse disciplines and curious newcomers to the field. The workshop embraces a broad definition of materials design encompassing matter in various forms, such as crystalline and amorphous solid-state materials, glasses, molecules, nanomaterials, and devices. By taking a comprehensive look at automated materials discovery spanning AI-guided design, synthesis and automated material characterization, we hope to create an opportunity for deep, thoughtful discussion among researchers working on these interdisciplinary topics, and highlight ongoing challenges in the field.\n\nAI4Mat was first held at NeurIPS 2022, bringing together materials scientists and AI researchers into a common forum with productive discussion on major research challenges at the intersection of AI and materials science. Since then, AI4Mat has established itself as a leading venue for the exchange of ideas on the latest developments in the field, bridging together international academic, industry and government institutions. AI4Mat-NeurIPS-2023 highlighted the growing interest and expanding research community of this emerging field. This momentum continued with two workshops held in 2024 (AI4Mat-BOKU-2024 in Vienna and AI4Mat-NeurIPS-2024 in Vancouver) designed to further accelerate research progress. The field of AI-enabled materials discovery is increasingly propelled by a global and interdisciplinary research community, whose collaborative efforts are driving materials innovation toward tangible real-world impact across diverse applications.\n\nAI4Mat-ICLR-2025 in Singapore, AI4Mat's first workshop in Asia, focused on the role of foundation models and representation learning for materials science while continuing to build a more global community of researchers for the emerging field. AI4Mat-NeurIPS-2025 focused discussion on latest frontiers and approaches to benchmarking while introducing a new format of live feedback for selected papers through AI4Mat-RLSF (Research Learning from Speaker Feedback). AI4Mat-ICLR-2026, AI4Mat's first workshop in South America, continued growing the global community while hosting discussions on feedback-based learning and multi-modal representations. The AI4Mat-NeurIPS-2026 will continue this effort by further expanding the workshop's geographic reach and a new program will focus on:\n\nScaling Laws for Materials Reasoning: From Compute to Scientific Discovery:\nThe progress of foundation models has been driven by scaling laws relating model size, data, and compute to predictable capability gains. Materials science, however, presents a fundamentally different scaling landscape: data is heterogeneous, expensive, and success is often driven by deep thought and domain specific reasoning. In this session, we will explore how scaling laws show up in the materials domain, with a focus on reasoning-intensive tasks including but not limited to synthesis planning, property prediction under structural and compositional complexity, inverse design, reasoning model training and training of materials science specific models like Machine Learning Interatomic Potentials (MLIPs). As such, some of the motivating questions include: Do larger models yield predictable improvements on materials benchmarks, or do domain-specific bottlenecks break conventional scaling behavior? How can test-time compute and chain-of-thought reasoning be leveraged for problems requiring deep scientific reasoning? What are the returns on scaling compute for simulation-in-the-loop workflows where each data point carries significant cost?\n\nAutomating Discovery That Delivers: When AI Meets the Messiness of Real Experiments:\nGenerative models, universal potentials, and agentic design loops now populate the machine learning literature at an increasing pace, yet a persistent gap remains between algorithmic innovation and tangible deployment for materials discovery. In this session, we aim to directly address this gap, sharing firsthand research focusing on systems that deliver impactful discovery ranging from AI-driven prediction through robotic synthesis, automated characterization, and iterative refinement. We also aim to focus on automated characterization and data collection as a critical bottleneck in autonomous discovery, including high-throughput measurements that might generate rich but noisy data streams. This session will also explore practical challenges, such as distribution shifts, integration challenges with robotic platforms and self-driving laboratories, and the underlying infrastructure required for autonomous systems that deliver beyond the conference paper.\n\nSubmissions\nCheck our submissions page for instructions on how to submit through OpenReview. Accepted peer-reviewed submissions will be invited to present a poster at the workshop and posted on the workshop website for non-archival records. Some peer-reviewed submissions will be invited to present a spotlight talk.\n\nContact\nEmail: ai4mat@googlegroups.com", "cfp_status": "published", "topics": [ "Scaling laws for materials reasoning: from compute to scientific discovery", "Reasoning-intensive tasks: synthesis planning, property prediction, inverse design", "Reasoning model training and Machine Learning Interatomic Potentials (MLIPs)", "Test-time compute and chain-of-thought reasoning for scientific reasoning", "Simulation-in-the-loop workflows", "Automating discovery: generative models, universal potentials, agentic design loops", "AI-driven prediction through robotic synthesis", "Automated characterization and data collection", "Distribution shifts and integration with robotic platforms and self-driving laboratories", "Materials design across crystalline and amorphous solids, glasses, molecules, nanomaterials, and devices" ], "important_dates": [ { "label": "Submission Start", "date": "2026-08-28" }, { "label": "Submission Deadline", "date": "2026-08-30" } ], "organizers": [ "Santiago Miret (Lila Sciences)", "Mara Schilling-Wilhelmi (Friedrich Schiller University Jena)", "Vijay Narasimhan (Merck KGaA, Darmstadt, Germany)", "N M Anoop Krishnan (IIT Delhi)", "Stefano Martiniani (New York University)" ], "speakers": [], "host_url": "https://sites.google.com/view/ai4mat/home", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AI4Mat", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AI4Mat", "location": "Sydney, Australia", "city": "Sydney", "workshop_date": "", "contact": "ai4mat@googlegroups.com", "tracks": [ { "key": "AI4Mat", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AI4Mat", "submission_dates_raw": "Submission Start: Aug 28 2026 12:00PM UTC-0, Submission Deadline: Aug 30 2026 12:00PM UTC-0" } ], "group": "science", "group_label": "AI for Science", "corpus": "AI for Accelerated Materials Design - NeurIPS 2026 AI4Mat-NeurIPS-2026 An interdisciplinary workshop bringing AI researchers and materials scientists together to tackle challenges in AI-driven materials discovery and development, spanning AI-guided design, synthesis, and automated characterization. Scaling laws for materials reasoning: from compute to scientific discovery Reasoning-intensive tasks: synthesis planning, property prediction, inverse design Reasoning model training and Machine Learning Interatomic Potentials (MLIPs) Test-time compute and chain-of-thought reasoning for scientific reasoning Simulation-in-the-loop workflows Automating discovery: generative models, universal potentials, agentic design loops AI-driven prediction through robotic synthesis Automated characterization and data collection Distribution shifts and integration with robotic platforms and self-driving laboratories Materials design across crystalline and amorphous solids, glasses, molecules, nanomaterials, and devices AI4Mat: AI for Accelerated Materials Design\nDecember 2026 @ NeurIPS 2026 (Sydney, Australia)\n\nAbout the Workshop\nThe AI for Accelerated Materials Discovery (AI4Mat) Workshop at NeurIPS 2026 provides an inclusive and collaborative platform where AI researchers and material scientists converge to tackle the cutting-edge challenges in AI-driven materials discovery and development. Our goal is to foster a vibrant exchange of ideas, breaking down barriers between disciplines and encouraging insightful discussions among experts from diverse disciplines and curious newcomers to the field. The workshop embraces a broad definition of materials design encompassing matter in various forms, such as crystalline and amorphous solid-state materials, glasses, molecules, nanomaterials, and devices. By taking a comprehensive look at automated materials discovery spanning AI-guided design, synthesis and automated material characterization, we hope to create an opportunity for deep, thoughtful discussion among researchers working on these interdisciplinary topics, and highlight ongoing challenges in the field.\n\nAI4Mat was first held at NeurIPS 2022, bringing together materials scientists and AI researchers into a common forum with productive discussion on major research challenges at the intersection of AI and materials science. Since then, AI4Mat has established itself as a leading venue for the exchange of ideas on the latest developments in the field, bridging together international academic, industry and government institutions. AI4Mat-NeurIPS-2023 highlighted the growing interest and expanding research community of this emerging field. This momentum continued with two workshops held in 2024 (AI4Mat-BOKU-2024 in Vienna and AI4Mat-NeurIPS-2024 in Vancouver) designed to further accelerate research progress. The field of AI-enabled materials discovery is increasingly propelled by a global and interdisciplinary research community, whose collaborative efforts are driving materials innovation toward tangible real-world impact across diverse applications.\n\nAI4Mat-ICLR-2025 in Singapore, AI4Mat's first workshop in Asia, focused on the role of foundation models and representation learning for materials science while continuing to build a more global community of researchers for the emerging field. AI4Mat-NeurIPS-2025 focused discussion on latest frontiers and approaches to benchmarking while introducing a new format of live feedback for selected papers through AI4Mat-RLSF (Research Learning from Speaker Feedback). AI4Mat-ICLR-2026, AI4Mat's first workshop in South America, continued growing the global community while hosting discussions on feedback-based learning and multi-modal representations. The AI4Mat-NeurIPS-2026 will continue this effort by further expanding the workshop's geographic reach and a new program will focus on:\n\nScaling Laws for Materials Reasoning: From Compute to Scientific Discovery:\nThe progress of foundation models has been driven by scaling laws relating model size, data, and compute to predictable capability gains. Materials science, however, presents a fundamentally different scaling landscape: data is heterogeneous, expensive, and success is often driven by deep thought and domain specific reasoning. In this session, we will explore how scaling laws show up in the materials domain, with a focus on reasoning-intensive tasks including but not limited to synthesis planning, property prediction under structural and compositional complexity, inverse design, reasoning model training and training of materials science specific models like Machine Learning Interatomic Potentials (MLIPs). As such, some of the motivating questions include: Do larger models yield predictable improvements on materials benchmarks, or do domain-specific bottlenecks break conventional scaling behavior? How can test-time compute and chain-of-thought reasoning be leveraged for problems requiring deep scientific reasoning? What are the returns on scaling compute for simulation-in-the-loop workflows where each data point carries significant cost?\n\nAutomating Discovery That Delivers: When AI Meets the Messiness of Real Experiments:\nGenerative models, universal potentials, and agentic design loops now populate the machine learning literature at an increasing pace, yet a persistent gap remains between algorithmic innovation and tangible deployment for materials discovery. In this session, we aim to directly address this gap, sharing firsthand research focusing on systems that deliver impactful discovery ranging from AI-driven prediction through robotic synthesis, automated characterization, and iterative refinement. We also aim to focus on automated characterization and data collection as a critical bottleneck in autonomous discovery, including high-throughput measurements that might generate rich but noisy data streams. This session will also explore practical challenges, such as distribution shifts, integration challenges with robotic platforms and self-driving laboratories, and the underlying infrastructure required for autonomous systems that deliver beyond the conference paper.\n\nSubmissions\nCheck our submissions page for instructions on how to submit through OpenReview. Accepted peer-reviewed submissions will be invited to present a poster at the workshop and posted on the workshop website for non-archival records. Some peer-reviewed submissions will be invited to present a spotlight talk.\n\nContact\nEmail: ai4mat@googlegroups.com" }, { "key": "AI4ChipDesign", "title": "AI for Chip Design - NeurIPS 2026 Workshop", "subtitle": "AI4ChipDesign 2026", "summary": "AI for Chip Design brings together researchers and practitioners at the intersection of artificial intelligence and semiconductor design, exploring machine learning methods for design automation, optimization, verification, and hardware-aware learning across the ML, electronic design automation, and hardware communities.", "cfp_full": "AI for Chip Design\nNeurIPS 2026 Workshop\nDate & Location: December 2026, Paris, France.\n\nCall for Papers\nSubmit your latest research in AI-driven chip design.\n\nThis workshop brings together researchers and practitioners working at the intersection of artificial intelligence and semiconductor design.\n\nThe workshop will explore advances in machine learning methods for chip design, including emerging AI-driven approaches for design automation, optimization, verification, and hardware-aware learning.\n\nThe goal is to foster discussion between the machine learning, electronic design automation, and hardware communities, and to identify new opportunities for AI-enabled chip development.\n\nImportant Dates\n- Paper & Poster Submission: August 30, 2026 (AoE)\n- Notification: September 29, 2026 (AoE)\n- Camera-ready Deadline: October 9, 2026 (AoE)\n- Workshop: December 12th, 2026, Paris\n\nNews\n- July 26th 2026 - Submission site open\n- July 22nd 2026 - Call for Papers published\n- July 12th 2026 - Workshop accepted at NeurIPS 2026, Paris venue.\n\nSubmission\nSubmissions are now open! You can submit your work via OpenReview.\n\nTopics of Interest\nTopics include, but are not limited to:\n- Machine learning for physical design: placement, routing, and floorplanning\n- ML for RTL, logic synthesis, and technology mapping\n- Timing, power, and area prediction and optimization\n- Graph neural networks for circuits and netlists\n- Generative models (e.g., diffusion, flow matching) for layout and design\n- Reinforcement learning for EDA tasks, including script generation, testbench production, and module completion\n- Large language models and foundation models for hardware description languages (Verilog, VHDL, and SystemVerilog)\n- Agentic systems for chip design workflows with tool use (linters, synthesizers, simulators, timing analyzers, etc.)\n- Reproducible benchmarks, datasets, and open-source infrastructure\n- Societal, educational, and workforce aspects of AI-driven chip design\n\nWe particularly encourage submissions that promote reproducibility through open-source code, datasets, model weights, evaluation frameworks, or other openly available research artifacts.\n\nContact\nFor questions, please contact: neurips-ai-chip-design-2026@bsc.es", "cfp_status": "published", "topics": [ "Machine learning for physical design: placement, routing, and floorplanning", "ML for RTL, logic synthesis, and technology mapping", "Timing, power, and area prediction and optimization", "Graph neural networks for circuits and netlists", "Generative models (e.g., diffusion, flow matching) for layout and design", "Reinforcement learning for EDA tasks, including script generation, testbench production, and module completion", "Large language models and foundation models for hardware description languages (Verilog, VHDL, and SystemVerilog)", "Agentic systems for chip design workflows with tool use (linters, synthesizers, simulators, timing analyzers, etc.)", "Reproducible benchmarks, datasets, and open-source infrastructure", "Societal, educational, and workforce aspects of AI-driven chip design" ], "important_dates": [ { "label": "Paper & Poster Submission (AoE)", "date": "2026-08-30" }, { "label": "Notification (AoE)", "date": "2026-09-29" }, { "label": "Camera-ready Deadline (AoE)", "date": "2026-10-09" }, { "label": "Workshop", "date": "2026-12-12" } ], "organizers": [], "speakers": [], "host_url": "https://neurips-ai-for-chip-design-2026.github.io/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AI4ChipDesign", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AI4ChipDesign", "location": "Paris, France", "city": "Paris", "workshop_date": "2026-12-12", "contact": "neurips-ai-chip-design-2026@bsc.es", "tracks": [ { "key": "AI4ChipDesign", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AI4ChipDesign", "submission_dates_raw": "" } ], "group": "science", "group_label": "AI for Science", "corpus": "AI for Chip Design - NeurIPS 2026 Workshop AI4ChipDesign 2026 AI for Chip Design brings together researchers and practitioners at the intersection of artificial intelligence and semiconductor design, exploring machine learning methods for design automation, optimization, verification, and hardware-aware learning across the ML, electronic design automation, and hardware communities. Machine learning for physical design: placement, routing, and floorplanning ML for RTL, logic synthesis, and technology mapping Timing, power, and area prediction and optimization Graph neural networks for circuits and netlists Generative models (e.g., diffusion, flow matching) for layout and design Reinforcement learning for EDA tasks, including script generation, testbench production, and module completion Large language models and foundation models for hardware description languages (Verilog, VHDL, and SystemVerilog) Agentic systems for chip design workflows with tool use (linters, synthesizers, simulators, timing analyzers, etc.) Reproducible benchmarks, datasets, and open-source infrastructure Societal, educational, and workforce aspects of AI-driven chip design AI for Chip Design\nNeurIPS 2026 Workshop\nDate & Location: December 2026, Paris, France.\n\nCall for Papers\nSubmit your latest research in AI-driven chip design.\n\nThis workshop brings together researchers and practitioners working at the intersection of artificial intelligence and semiconductor design.\n\nThe workshop will explore advances in machine learning methods for chip design, including emerging AI-driven approaches for design automation, optimization, verification, and hardware-aware learning.\n\nThe goal is to foster discussion between the machine learning, electronic design automation, and hardware communities, and to identify new opportunities for AI-enabled chip development.\n\nImportant Dates\n- Paper & Poster Submission: August 30, 2026 (AoE)\n- Notification: September 29, 2026 (AoE)\n- Camera-ready Deadline: October 9, 2026 (AoE)\n- Workshop: December 12th, 2026, Paris\n\nNews\n- July 26th 2026 - Submission site open\n- July 22nd 2026 - Call for Papers published\n- July 12th 2026 - Workshop accepted at NeurIPS 2026, Paris venue.\n\nSubmission\nSubmissions are now open! You can submit your work via OpenReview.\n\nTopics of Interest\nTopics include, but are not limited to:\n- Machine learning for physical design: placement, routing, and floorplanning\n- ML for RTL, logic synthesis, and technology mapping\n- Timing, power, and area prediction and optimization\n- Graph neural networks for circuits and netlists\n- Generative models (e.g., diffusion, flow matching) for layout and design\n- Reinforcement learning for EDA tasks, including script generation, testbench production, and module completion\n- Large language models and foundation models for hardware description languages (Verilog, VHDL, and SystemVerilog)\n- Agentic systems for chip design workflows with tool use (linters, synthesizers, simulators, timing analyzers, etc.)\n- Reproducible benchmarks, datasets, and open-source infrastructure\n- Societal, educational, and workforce aspects of AI-driven chip design\n\nWe particularly encourage submissions that promote reproducibility through open-source code, datasets, model weights, evaluation frameworks, or other openly available research artifacts.\n\nContact\nFor questions, please contact: neurips-ai-chip-design-2026@bsc.es" }, { "key": "AI4PowerGrids", "title": "AI Foundations for Power Grids @ NeurIPS 2026", "subtitle": "AI4PowerGrids 2026", "summary": "A NeurIPS 2026 workshop bringing power systems to the ML community as a first-class methodological challenge, focused on benchmarks, model training, and how to evaluate learning-based methods for the power grid under realistic and evolving operating conditions.", "cfp_full": "About the workshop\nWhat would it take for AI models to enter a power grid control room? A workshop on benchmarks, model training, and real-world considerations.\n\nThe power grid is one of the most consequential open problems in applied machine learning – hard physics constraints, real-time closed-loop operation, structural non-stationarity, and societal-scale consequence – yet it attracts a fraction of the methodological attention given to vision, language, or biology. This workshop brings power systems to the ML community as a first-class methodological challenge and focuses on the central bottleneck: how to evaluate learning-based methods under realistic and evolving operating conditions.\n\nKey dates (AoE)\n- Submission deadline: August 29, 2026\n- Author notification: September 29, 2026\n- Workshop: December 11 or 12, 2026 (Sydney)\n\nSubmission tracks\n- Methods with rigorous evaluation.\n- Benchmarks, datasets, and evaluation protocols.\n- Position and empirical-evaluation papers.\n- Negative results and failure modes.\n\nThe workshop especially welcomes contributions on open, realistic datasets as a first-class community deliverable, beyond legacy IEEE test cases; physics-respecting metrics for stochastic models; structural shifts in grid topology and generation patterns; foundation-model claims and their evaluation criteria; component accuracy versus system-level stability; and the progression from supervised to autonomous operation.\n\nDomain evaluation checklist\nAll submissions must include a domain checklist verifying:\n- Physics feasibility under full AC power-flow equations, with constraint-violation statistics\n- Out-of-distribution evaluation across unseen topologies or operating regimes\n- Documented failure cases and tail statistics\n- System-level downstream effects for embedded components\n- Data and code availability plans\n\nSubmission format\nUp to 4 pages of main content, plus unlimited references. Submissions are made via OpenReview and receive at least 3 double-blind reviews. Original work unpublished at ML venues qualifies; prior power-systems publication is acceptable if explicitly disclosed. Position papers and work-in-progress are welcome. The workshop is non-archival with no formal proceedings.\n\nSubmit\nSubmit via OpenReview: openreview.net/group?id=NeurIPS.cc/2026/Workshop/AI4PowerGrids", "cfp_status": "published", "topics": [ "Methods with rigorous evaluation", "Benchmarks, datasets, and evaluation protocols", "Position and empirical-evaluation papers", "Negative results and failure modes", "Open, realistic datasets beyond legacy IEEE test cases", "Physics-respecting metrics for stochastic models", "Structural shifts in grid topology and generation patterns", "Foundation-model claims and their evaluation criteria", "Component accuracy versus system-level stability", "From supervised to autonomous operation" ], "important_dates": [ { "label": "Submission Deadline", "date": "2026-08-29" }, { "label": "Author Notification", "date": "2026-09-29" }, { "label": "Workshop", "date": "2026-12-11 or 2026-12-12" } ], "organizers": [ "Andrea Britto Mattos Lima (Microsoft Research)", "Thomas Brunschwiler (IBM Research)", "Nicolas Christianson (Johns Hopkins University)", "Wenqi Cui (NYU)", "Rabab Haider (University of Michigan)", "Christopher Yeh (Harvard)", "Baosen Zhang (University of Washington)" ], "speakers": [], "host_url": "https://ai4powergrids.github.io", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AI4PowerGrids", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AI4PowerGrids", "location": "Sydney, Australia", "city": "Sydney", "workshop_date": "2026-12-10", "contact": "ai4powergrids@gmail.com", "tracks": [ { "key": "AI4PowerGrids", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AI4PowerGrids", "submission_dates_raw": "" } ], "group": "science", "group_label": "AI for Science", "corpus": "AI Foundations for Power Grids @ NeurIPS 2026 AI4PowerGrids 2026 A NeurIPS 2026 workshop bringing power systems to the ML community as a first-class methodological challenge, focused on benchmarks, model training, and how to evaluate learning-based methods for the power grid under realistic and evolving operating conditions. Methods with rigorous evaluation Benchmarks, datasets, and evaluation protocols Position and empirical-evaluation papers Negative results and failure modes Open, realistic datasets beyond legacy IEEE test cases Physics-respecting metrics for stochastic models Structural shifts in grid topology and generation patterns Foundation-model claims and their evaluation criteria Component accuracy versus system-level stability From supervised to autonomous operation About the workshop\nWhat would it take for AI models to enter a power grid control room? A workshop on benchmarks, model training, and real-world considerations.\n\nThe power grid is one of the most consequential open problems in applied machine learning – hard physics constraints, real-time closed-loop operation, structural non-stationarity, and societal-scale consequence – yet it attracts a fraction of the methodological attention given to vision, language, or biology. This workshop brings power systems to the ML community as a first-class methodological challenge and focuses on the central bottleneck: how to evaluate learning-based methods under realistic and evolving operating conditions.\n\nKey dates (AoE)\n- Submission deadline: August 29, 2026\n- Author notification: September 29, 2026\n- Workshop: December 11 or 12, 2026 (Sydney)\n\nSubmission tracks\n- Methods with rigorous evaluation.\n- Benchmarks, datasets, and evaluation protocols.\n- Position and empirical-evaluation papers.\n- Negative results and failure modes.\n\nThe workshop especially welcomes contributions on open, realistic datasets as a first-class community deliverable, beyond legacy IEEE test cases; physics-respecting metrics for stochastic models; structural shifts in grid topology and generation patterns; foundation-model claims and their evaluation criteria; component accuracy versus system-level stability; and the progression from supervised to autonomous operation.\n\nDomain evaluation checklist\nAll submissions must include a domain checklist verifying:\n- Physics feasibility under full AC power-flow equations, with constraint-violation statistics\n- Out-of-distribution evaluation across unseen topologies or operating regimes\n- Documented failure cases and tail statistics\n- System-level downstream effects for embedded components\n- Data and code availability plans\n\nSubmission format\nUp to 4 pages of main content, plus unlimited references. Submissions are made via OpenReview and receive at least 3 double-blind reviews. Original work unpublished at ML venues qualifies; prior power-systems publication is acceptable if explicitly disclosed. Position papers and work-in-progress are welcome. The workshop is non-archival with no formal proceedings.\n\nSubmit\nSubmit via OpenReview: openreview.net/group?id=NeurIPS.cc/2026/Workshop/AI4PowerGrids" }, { "key": "AIDaR", "title": "AIDaR: AI Data Readiness for Scientific Discovery", "subtitle": "AIDaR @ NeurIPS 2026", "summary": "A one-day NeurIPS 2026 workshop on data infrastructure and benchmarks for reliable scientific AI systems, focused on how scientific data should be organized so models and agents can use it, and how evaluations should determine where frontier models are reliable.", "cfp_full": "AIDaR: AI Data Readiness for Scientific Discovery\n\nData Infrastructure and Benchmarks for Reliable Scientific AI Systems. One-day in-person workshop. Paris, France. December 12 or 13, 2026.\n\nOverview\n\nScientific AI is moving from curated prediction tasks toward systems that work directly with experimental data: foundation models, retrieval systems, and agents that query scientific knowledge, run analyses, and support follow-up decisions.\n\nAIDaR focuses on two problems in this transition — how scientific data should be organized so models and agents can use it, and how evaluations should determine where frontier models are reliable. The workshop brings together researchers, industry scientists, and engineers building scientific data systems, foundation models, agents, biomedical evaluations, and industrial assay platforms.\n\nWorkshop details\n\nData infrastructure for scientific AI\nScientific data must move from raw measurements into forms that models, retrieval systems, and agents can use: linking files to samples and assay conditions, preserving the steps that produced an analysis-ready dataset, and representing scientific relationships as tables, graphs, or knowledge networks. Topics include scientific foundation models, relational and graph-structured data, multimodal biomedical measurements, workflow systems, and large industrial datasets used for model training and wet-lab feedback.\n\nEvaluations for scientific AI\nScientific AI systems should be tested on tasks that resemble real scientific work: choosing inputs, running analyses, checking controls, interpreting noisy results, and recovering biological or physical conclusions from data. Topics include biological reasoning, practical data analysis, retrieval over structured scientific knowledge, multimodal biomedical data, failure analysis, and studies of where performance changes across assays, platforms, or workflows.\n\nFormat\nStructure. Short invited talks, central contributed work, poster and demo sessions, and breakout groups that produce short written outputs for the post-workshop report.\nTalks. Invited talks open with scientific data systems and ground the evaluation discussions in biomedical data.\nPanels. Panel 1 asks how to build evaluations that approximate real scientific work rather than ranking models on static benchmarks. Panel 2 covers the infrastructure that makes experimental data usable by scientific AI systems.\nBreakouts. Groups will examine real use-cases, considering how scientific datasets, metadata, and human feedback should be organized for model and agent use, as well as the conceptual framing of the AI data readiness problem. Amongst confirmed session leaders are Brandon White (Axiom), Fabio Boniolo and Matthew Osman (Polyphron).\n\nCall for papers\n\nWe invite technical papers, benchmarks and evaluations, data-system reports, workflow and tool demos, and failure analyses. Submissions should connect data organization or evaluation design to downstream model or agent behavior on scientific tasks.\n\nSubmission types\n- Full papers (up to 8 pages) present complete work: a scientific data system, benchmark, or evaluation, with methods, results, and analysis.\n- Short papers (up to 4 pages) present focused contributions, work in progress, position pieces, or tool and dataset demos.\n\nFormatting and review\n- Papers must use the NeurIPS 2026 style files; references and appendices are excluded from the page limit.\n- Reviewing is double-blind; anonymize the submission and any linked material.\n- Disclose any use of large language models and their role.\n- The workshop is non-archival.\n- OpenReview is the official submission system. You may optionally add a checked, machine-readable copy of your project through AIDaRS, our GitHub review pilot.\n\nKey dates:\n- Submission: August 29, 2026, 11:59 p.m. AoE\n- Notification: September 29, 2026, 11:59 p.m. AoE\n- Camera-ready: To be announced\n- Workshop: December 12 or 13, 2026, Paris, France", "cfp_status": "published", "topics": [ "Scientific foundation models", "Relational and graph-structured data", "Multimodal biomedical measurements", "Workflow systems and large industrial datasets for training and wet-lab feedback", "Biological reasoning and practical data analysis", "Retrieval over structured scientific knowledge", "Failure analysis and reliability across assays, platforms, or workflows", "Data infrastructure for scientific AI", "Evaluations for scientific AI" ], "important_dates": [ { "label": "Submission", "date": "2026-08-29" }, { "label": "Notification", "date": "2026-09-29" }, { "label": "Camera-ready", "date": "To be announced" }, { "label": "Workshop", "date": "2026-12-12" } ], "organizers": [ "Michal Rosen-Zvi (Hebrew University / Merck KGaA)", "Zoe Piran (Stanford / Roche / Genentech)", "Kenny Workman (LatchBio)", "Edaeni Hamid (Roche / Genentech)", "Vladimir Ermakov (Edison Scientific)", "Arindam Sett (Roche / Genentech)", "Sina Booeshaghi (UC Berkeley)", "Aviv Regev (Roche / Genentech) — Advisory Board", "Sam Rodriques (Edison Scientific) — Advisory Board", "Mohsen Hejrati (Apple) — Advisory Board", "Arvind Rajpal (Merck KGaA) — Advisory Board", "Lior Pachter (Caltech) — Advisory Board" ], "speakers": [ "Max Welling (University of Amsterdam)", "Marianna Rapsomaniki (University of Lausanne)", "Arjun Raj (Cellular Intelligence)", "Xinyi Zhang (Aithyra)", "Harihara Muralidharan (LatchBio) — Panelist", "Kexin Huang (Phylo / Biomni) — Panelist", "Karin Hrovatin (Merck KGaA) — Panelist", "Pablo Meyer Rojas (IBM Research / DREAM) — Panelist", "Jon Laurent (Edison Scientific) — Panelist", "Stefan Harrer (Sanofi) — Panelist" ], "host_url": "https://aidar-workshop.github.io/2026/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AIDaR", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AIDaR", "location": "Paris, France", "city": "Paris", "workshop_date": "2026-12-12", "contact": "aidar.neurips.2026.paris@gmail.com", "tracks": [ { "key": "AIDaR", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AIDaR", "submission_dates_raw": "" } ], "group": "science", "group_label": "AI for Science", "corpus": "AIDaR: AI Data Readiness for Scientific Discovery AIDaR @ NeurIPS 2026 A one-day NeurIPS 2026 workshop on data infrastructure and benchmarks for reliable scientific AI systems, focused on how scientific data should be organized so models and agents can use it, and how evaluations should determine where frontier models are reliable. Scientific foundation models Relational and graph-structured data Multimodal biomedical measurements Workflow systems and large industrial datasets for training and wet-lab feedback Biological reasoning and practical data analysis Retrieval over structured scientific knowledge Failure analysis and reliability across assays, platforms, or workflows Data infrastructure for scientific AI Evaluations for scientific AI AIDaR: AI Data Readiness for Scientific Discovery\n\nData Infrastructure and Benchmarks for Reliable Scientific AI Systems. One-day in-person workshop. Paris, France. December 12 or 13, 2026.\n\nOverview\n\nScientific AI is moving from curated prediction tasks toward systems that work directly with experimental data: foundation models, retrieval systems, and agents that query scientific knowledge, run analyses, and support follow-up decisions.\n\nAIDaR focuses on two problems in this transition — how scientific data should be organized so models and agents can use it, and how evaluations should determine where frontier models are reliable. The workshop brings together researchers, industry scientists, and engineers building scientific data systems, foundation models, agents, biomedical evaluations, and industrial assay platforms.\n\nWorkshop details\n\nData infrastructure for scientific AI\nScientific data must move from raw measurements into forms that models, retrieval systems, and agents can use: linking files to samples and assay conditions, preserving the steps that produced an analysis-ready dataset, and representing scientific relationships as tables, graphs, or knowledge networks. Topics include scientific foundation models, relational and graph-structured data, multimodal biomedical measurements, workflow systems, and large industrial datasets used for model training and wet-lab feedback.\n\nEvaluations for scientific AI\nScientific AI systems should be tested on tasks that resemble real scientific work: choosing inputs, running analyses, checking controls, interpreting noisy results, and recovering biological or physical conclusions from data. Topics include biological reasoning, practical data analysis, retrieval over structured scientific knowledge, multimodal biomedical data, failure analysis, and studies of where performance changes across assays, platforms, or workflows.\n\nFormat\nStructure. Short invited talks, central contributed work, poster and demo sessions, and breakout groups that produce short written outputs for the post-workshop report.\nTalks. Invited talks open with scientific data systems and ground the evaluation discussions in biomedical data.\nPanels. Panel 1 asks how to build evaluations that approximate real scientific work rather than ranking models on static benchmarks. Panel 2 covers the infrastructure that makes experimental data usable by scientific AI systems.\nBreakouts. Groups will examine real use-cases, considering how scientific datasets, metadata, and human feedback should be organized for model and agent use, as well as the conceptual framing of the AI data readiness problem. Amongst confirmed session leaders are Brandon White (Axiom), Fabio Boniolo and Matthew Osman (Polyphron).\n\nCall for papers\n\nWe invite technical papers, benchmarks and evaluations, data-system reports, workflow and tool demos, and failure analyses. Submissions should connect data organization or evaluation design to downstream model or agent behavior on scientific tasks.\n\nSubmission types\n- Full papers (up to 8 pages) present complete work: a scientific data system, benchmark, or evaluation, with methods, results, and analysis.\n- Short papers (up to 4 pages) present focused contributions, work in progress, position pieces, or tool and dataset demos.\n\nFormatting and review\n- Papers must use the NeurIPS 2026 style files; references and appendices are excluded from the page limit.\n- Reviewing is double-blind; anonymize the submission and any linked material.\n- Disclose any use of large language models and their role.\n- The workshop is non-archival.\n- OpenReview is the official submission system. You may optionally add a checked, machine-readable copy of your project through AIDaRS, our GitHub review pilot.\n\nKey dates:\n- Submission: August 29, 2026, 11:59 p.m. AoE\n- Notification: September 29, 2026, 11:59 p.m. AoE\n- Camera-ready: To be announced\n- Workshop: December 12 or 13, 2026, Paris, France" }, { "key": "AXIOM", "title": "AXIOM: Foundations of Efficient Deep Learning", "subtitle": "AXIOM", "summary": "A workshop bringing together researchers from deep learning theory, ML systems, optimization, and efficient AI to explore whether predictive principles can guide the design of efficient AI systems under real-world constraints of compute, memory, energy, communication, and data.", "cfp_full": "About the Workshop\n\nRecent advances in deep learning theory suggest that machine learning is gradually evolving from an empirical discipline into a predictive science. Scaling laws, optimization theory, learning dynamics, and representation learning increasingly explain why modern deep learning systems work. Yet these advances have had limited impact on one of today's most pressing challenges: building AI systems that are efficient under realistic constraints of compute, memory, energy, communication, and data. Current efficiency techniques, such as pruning, quantization, sparse computation, adaptive inference, and efficient architecture, remain largely driven by empirical investigation. Conversely, many theoretical advances explain observations only after the fact, rather than predicting which algorithms or architectures will be most effective before expensive experimentation.\n\nThe AXIOM Workshop brings together researchers from deep learning theory, machine learning systems, optimization, and efficient AI to explore a central question: Can we develop predictive principles that guide the design of efficient AI systems?\n\nOur goal is to bridge theory and practice by identifying the mathematical principles underlying efficient learning and by defining the next generation of research challenges for efficient deep learning. The workshop features invited vision talks, contributed papers, posters, panel discussions, and a community-driven Grand Challenges initiative that will collectively shape a research agenda for the foundations of efficient AI.\n\nCall For Papers\n\nTopics of Interest\n\nWe welcome submissions addressing theoretical, algorithmic, and systems aspects of efficient deep learning, including but not limited to:\n- Predictive Theory for Efficient Learning: Scaling laws for efficient models, compute-optimal training and inference, predicting capability under resource constraints, optimization and learning dynamics, generalization under limited compute or data, phase transitions in efficient learning;\n- Sparsity, Compression, and Model Structure: Foundations of pruning and quantization, sparse and modular neural networks, lottery tickets and subnetworks, adaptive computation, neural architecture design, representation learning for efficiency;\n- Efficient Foundation Models: Efficient LLMs and multimodal models, efficient reasoning and adaptive inference, mixture-of-experts and modular architectures, test-time adaptation, memory-efficient training and inference, distillation and compression;\n- Foundations and Future Directions: Theoretical limits of efficient AI, new efficiency metrics and benchmarks, predictive models of training dynamics, interpretability of efficient models, mathematical foundations of efficient deep learning, emerging theoretical paradigms for efficient AI.\n\nInterdisciplinary work connecting theory, algorithms, systems, and hardware is particularly encouraged.\n\nPaper Submission\n\nWe invite submissions presenting original research, preliminary results, novel ideas, and emerging research directions that advance the theoretical and practical foundations of efficient deep learning. We particularly encourage work that connects deep learning theory with efficient machine learning, including studies that improve our understanding of resource-efficient learning, reveal new theoretical principles, or bridge the gap between mathematical foundations and practical AI systems.\n\nWe invite short paper submissions (4 pages, excluding references) which follow the NeurIPS workshop formatting guidelines. This workshop is non-archival. Outstanding submissions will be invited for a 15-minute oral presentation. All accepted papers will be presented during the poster session and published on the workshop website. A Best Paper Award will be presented during the workshop.\n\nGrand Challenges Track\n\nBeyond research papers, AXIOM introduces a Grand Challenges track designed to identify the most important open questions for the future of efficient AI. Instead of reporting completed research, Grand Challenges submissions should articulate important unanswered questions, theoretical gaps, surprising empirical observations, or future research opportunities. We particularly encourage contributions on:\n- Open theoretical questions motivated by efficient AI;\n- Efficiency methods lacking theoretical explanations;\n- Contradictions between theory and practice;\n- Missing benchmarks, evaluation methodologies, or efficiency metrics.\n\nAccepted submissions will be published on the workshop website and synthesized into a community report that will serve as the starting point for an interactive discussion session during the workshop. The insights from this discussion will contribute to a community position paper outlining a research agenda for predictive and efficient AI.\n\nGrand Challenges submissions should consist of a one-page abstract plus references, with no appendix. They will undergo a light review and accepted submissions will be published on the workshop webpage.\n\nImportant Dates\nPaper submission: August 29, 2026 (11:59 PM UTC-0)\nAuthor notification: September 29, 2026 (AoE)\nCamera-ready deadline: TBA\n\nSubmission Details\nSubmission site: OpenReview.\nSubmission format instructions:\n- Papers: 4 pages excluding references and appendix, double-blind;\n- Grand Challenges: 1 page excluding references, no appendix, double-blind.\nPlease use the NeurIPS 2026 LaTeX template. All papers will undergo peer review by the Program Committee.\n\nFollowing the Guidance for NeurIPS Workshop Proposals 2026, workshop submissions must not duplicate work previously published at machine learning or related conferences. Work presented at the main NeurIPS conference must not also appear in the workshop, including as part of an invited talk. All authors, reviewers, presenters, and participants are expected to follow the NeurIPS Code of Conduct.\n\nWorkshop Participation Policy\nThe workshop is intended to be an in-person interactive event. Accepted papers are therefore expected to be presented in person by at least one author or designated presenter. Remote oral presentations may be permitted only in exceptional circumstances (e.g., participation from another official NeurIPS location, visa or medical issues), subject to the workshop's technical constraints. If no presenter can attend in person, the paper will remain accepted but will be identified as \"not presented\" on the workshop website.", "cfp_status": "published", "topics": [ "Predictive Theory for Efficient Learning: scaling laws for efficient models, compute-optimal training and inference, predicting capability under resource constraints, optimization and learning dynamics, generalization under limited compute or data, phase transitions in efficient learning", "Sparsity, Compression, and Model Structure: foundations of pruning and quantization, sparse and modular neural networks, lottery tickets and subnetworks, adaptive computation, neural architecture design, representation learning for efficiency", "Efficient Foundation Models: efficient LLMs and multimodal models, efficient reasoning and adaptive inference, mixture-of-experts and modular architectures, test-time adaptation, memory-efficient training and inference, distillation and compression", "Foundations and Future Directions: theoretical limits of efficient AI, new efficiency metrics and benchmarks, predictive models of training dynamics, interpretability of efficient models, mathematical foundations of efficient deep learning, emerging theoretical paradigms for efficient AI" ], "important_dates": [ { "label": "Paper submission", "date": "2026-08-29" }, { "label": "Author notification", "date": "2026-09-29" }, { "label": "Camera-ready deadline", "date": "TBA" } ], "organizers": [ "Olga Saukh (Graz University of Technology, Austria)", "Linara Adilova (TU Dortmund and RC Trust, Germany)", "Bernhard Geiger (Graz University of Technology, Austria)", "Yedi Zhang (University College London, UK)", "Flavio Martinelli (EPFL Lausanne, Switzerland)" ], "speakers": [ "Rahim Entezari (ex-StabilityAI, ex-Wayve, UK)", "Katharina Eggensperger (Lamarr Institute and TU Dortmund, Germany)", "Michael Kamp (Lamarr Institute and TU Dortmund, Germany)", "Bruno Loureiro (CNRS & Ecole Normale Superieure, France)", "Alexander van Meegen (RWTH Aachen, Germany)", "Hannah Pinson (Eindhoven University of Technology (TU/e), Netherlands)" ], "host_url": "https://axiom-neurips2026.github.io/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AXIOM", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AXIOM", "location": "Paris", "city": "Paris", "workshop_date": "2026-12-12", "contact": "axiom.neurips2026@gmail.com", "tracks": [ { "key": "AXIOM", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AXIOM", "submission_dates_raw": "Submission Deadline: Aug 29 2026 11:59PM UTC-0" } ], "group": "fm", "group_label": "Foundation Models & LLMs", "corpus": "AXIOM: Foundations of Efficient Deep Learning AXIOM A workshop bringing together researchers from deep learning theory, ML systems, optimization, and efficient AI to explore whether predictive principles can guide the design of efficient AI systems under real-world constraints of compute, memory, energy, communication, and data. Predictive Theory for Efficient Learning: scaling laws for efficient models, compute-optimal training and inference, predicting capability under resource constraints, optimization and learning dynamics, generalization under limited compute or data, phase transitions in efficient learning Sparsity, Compression, and Model Structure: foundations of pruning and quantization, sparse and modular neural networks, lottery tickets and subnetworks, adaptive computation, neural architecture design, representation learning for efficiency Efficient Foundation Models: efficient LLMs and multimodal models, efficient reasoning and adaptive inference, mixture-of-experts and modular architectures, test-time adaptation, memory-efficient training and inference, distillation and compression Foundations and Future Directions: theoretical limits of efficient AI, new efficiency metrics and benchmarks, predictive models of training dynamics, interpretability of efficient models, mathematical foundations of efficient deep learning, emerging theoretical paradigms for efficient AI About the Workshop\n\nRecent advances in deep learning theory suggest that machine learning is gradually evolving from an empirical discipline into a predictive science. Scaling laws, optimization theory, learning dynamics, and representation learning increasingly explain why modern deep learning systems work. Yet these advances have had limited impact on one of today's most pressing challenges: building AI systems that are efficient under realistic constraints of compute, memory, energy, communication, and data. Current efficiency techniques, such as pruning, quantization, sparse computation, adaptive inference, and efficient architecture, remain largely driven by empirical investigation. Conversely, many theoretical advances explain observations only after the fact, rather than predicting which algorithms or architectures will be most effective before expensive experimentation.\n\nThe AXIOM Workshop brings together researchers from deep learning theory, machine learning systems, optimization, and efficient AI to explore a central question: Can we develop predictive principles that guide the design of efficient AI systems?\n\nOur goal is to bridge theory and practice by identifying the mathematical principles underlying efficient learning and by defining the next generation of research challenges for efficient deep learning. The workshop features invited vision talks, contributed papers, posters, panel discussions, and a community-driven Grand Challenges initiative that will collectively shape a research agenda for the foundations of efficient AI.\n\nCall For Papers\n\nTopics of Interest\n\nWe welcome submissions addressing theoretical, algorithmic, and systems aspects of efficient deep learning, including but not limited to:\n- Predictive Theory for Efficient Learning: Scaling laws for efficient models, compute-optimal training and inference, predicting capability under resource constraints, optimization and learning dynamics, generalization under limited compute or data, phase transitions in efficient learning;\n- Sparsity, Compression, and Model Structure: Foundations of pruning and quantization, sparse and modular neural networks, lottery tickets and subnetworks, adaptive computation, neural architecture design, representation learning for efficiency;\n- Efficient Foundation Models: Efficient LLMs and multimodal models, efficient reasoning and adaptive inference, mixture-of-experts and modular architectures, test-time adaptation, memory-efficient training and inference, distillation and compression;\n- Foundations and Future Directions: Theoretical limits of efficient AI, new efficiency metrics and benchmarks, predictive models of training dynamics, interpretability of efficient models, mathematical foundations of efficient deep learning, emerging theoretical paradigms for efficient AI.\n\nInterdisciplinary work connecting theory, algorithms, systems, and hardware is particularly encouraged.\n\nPaper Submission\n\nWe invite submissions presenting original research, preliminary results, novel ideas, and emerging research directions that advance the theoretical and practical foundations of efficient deep learning. We particularly encourage work that connects deep learning theory with efficient machine learning, including studies that improve our understanding of resource-efficient learning, reveal new theoretical principles, or bridge the gap between mathematical foundations and practical AI systems.\n\nWe invite short paper submissions (4 pages, excluding references) which follow the NeurIPS workshop formatting guidelines. This workshop is non-archival. Outstanding submissions will be invited for a 15-minute oral presentation. All accepted papers will be presented during the poster session and published on the workshop website. A Best Paper Award will be presented during the workshop.\n\nGrand Challenges Track\n\nBeyond research papers, AXIOM introduces a Grand Challenges track designed to identify the most important open questions for the future of efficient AI. Instead of reporting completed research, Grand Challenges submissions should articulate important unanswered questions, theoretical gaps, surprising empirical observations, or future research opportunities. We particularly encourage contributions on:\n- Open theoretical questions motivated by efficient AI;\n- Efficiency methods lacking theoretical explanations;\n- Contradictions between theory and practice;\n- Missing benchmarks, evaluation methodologies, or efficiency metrics.\n\nAccepted submissions will be published on the workshop website and synthesized into a community report that will serve as the starting point for an interactive discussion session during the workshop. The insights from this discussion will contribute to a community position paper outlining a research agenda for predictive and efficient AI.\n\nGrand Challenges submissions should consist of a one-page abstract plus references, with no appendix. They will undergo a light review and accepted submissions will be published on the workshop webpage.\n\nImportant Dates\nPaper submission: August 29, 2026 (11:59 PM UTC-0)\nAuthor notification: September 29, 2026 (AoE)\nCamera-ready deadline: TBA\n\nSubmission Details\nSubmission site: OpenReview.\nSubmission format instructions:\n- Papers: 4 pages excluding references and appendix, double-blind;\n- Grand Challenges: 1 page excluding references, no appendix, double-blind.\nPlease use the NeurIPS 2026 LaTeX template. All papers will undergo peer review by the Program Committee.\n\nFollowing the Guidance for NeurIPS Workshop Proposals 2026, workshop submissions must not duplicate work previously published at machine learning or related conferences. Work presented at the main NeurIPS conference must not also appear in the workshop, including as part of an invited talk. All authors, reviewers, presenters, and participants are expected to follow the NeurIPS Code of Conduct.\n\nWorkshop Participation Policy\nThe workshop is intended to be an in-person interactive event. Accepted papers are therefore expected to be presented in person by at least one author or designated presenter. Remote oral presentations may be permitted only in exceptional circumstances (e.g., participation from another official NeurIPS location, visa or medical issues), subject to the workshop's technical constraints. If no presenter can attend in person, the paper will remain accepted but will be identified as \"not presented\" on the workshop website." }, { "key": "BeNTo", "title": "Beyond Next Token Prediction: Diffusion and Flow Models for Next-Generation Decoding", "subtitle": "BeNTo Workshop NeurIPS 2026", "summary": "A NeurIPS 2026 workshop exploring the theory, algorithms, applications, and systems of discrete diffusion and flow models for parallel, non-causal generation — moving beyond fixed-order autoregressive next-token prediction — across two tracks on foundations and on applications/systems.", "cfp_full": "Beyond Next-Token Prediction: Diffusion & Flow Models for Next-Generation Decoding\nBeyond next-token prediction — exploring the theory, algorithms, and applications of discrete diffusion and flow models for parallel, non-causal generation.\n\nOverview — Rethinking the order of generation.\nAutoregressive models produce their output one element at a time, in a fixed order. The recipe is remarkably general, and it has carried the field a long way. However, it also ties the order of computation to the order of the result, leaving parallelism, revision, and control hard to reach. Discrete diffusion and flow models take a different route. They generate by iterative denoising: refining many positions at once, and revisiting earlier choices as a sample takes shape.\n\nThat shift brings a different set of computational properties — non-causal, parallel, and controllable — along with open questions spanning mathematics, algorithms, and engineering, and a growing body of work that combines the two paradigms rather than choosing between them. This workshop joins two conversations: the theory and algorithms that make discrete generative models work, and the applications and systems that put them to use.\n\nTrack 01 — Depth: Theories & Algorithms\nDiscrete diffusion and flow models at the intersection of generative modeling, optimal transport, and stochastic optimal control.\n- Generative Modeling: Discrete diffusion & flow models, optimal transport, optimal control, Schrödinger bridges.\n- Probabilistic Inference: Discrete diffusion samplers, adjoint-based samplers, advanced MCMC, variational inference.\n\nTrack 02 — Breadth: Applications & Systems\nNovel applications and scalable systems that exploit non-causal, parallel generation.\n- Applications: AI for science, multimodal generation, reward alignment, inverse problems, benchmarks & datasets.\n- Systems & Empirical Analysis: Foundation models, large-scale training/inference, network architectures.\n\nScope — Topics of interest\nThis workshop considers, but is not limited to, the following topics.\n- Discrete diffusion & flow models\n- Optimal transport\n- Optimal & stochastic control\n- Schrödinger bridges\n- Discrete diffusion & adjoint-based samplers\n- MCMC & variational inference\n- AI for science\n- Multimodal generation\n- Reward alignment\n- Inverse problems\n- Benchmarks & datasets\n- Foundation models & large-scale training/inference\n- Architectures for non-autoregressive generation\n- Systems & hardware for diffusion models\n\nCall for Papers — Share your work.\nWhat to submit: We invite submissions on any topic within the workshop's scope, across both tracks, from theory and algorithms to applications and systems. All submissions are handled through OpenReview.\nPage limit: Submissions may be either 4 or 8 pages, excluding references and appendices.\nPeer review: Each submission receives 3 reviews; each reviewer handles at most 3 papers.\nMutual-review policy: At least one qualified author per submission serves as a reviewer.\nTravel grants for students and early-career researchers with accepted papers.\nPresentation: Accepted papers are presented as posters, with a selection of contributed orals (6 contributed orals, 10 min each). A Best Paper Award recognizes outstanding work.\nSubmissions are managed through OpenReview, at NeurIPS.cc/2026/Workshop/BeNTo. Questions? bento-neurips@googlegroups.com\n\nImportant Dates (all deadlines 23:59 AoE; dates indicative and subject to change)\n- Jul 20, 2026: Call for Papers & submissions open (OpenReview)\n- Aug 29, 2026: Paper submission deadline\n- Aug 30 – Sep 20, 2026: Peer review period\n- Sep 21 – 25, 2026: Meta-review & discussion\n- Sep 28, 2026: Accept / reject notification\n- Oct 20, 2026: Camera-ready deadline (tentative)\n- Dec 11 or 12, 2026: Workshop @ NeurIPS 2026 (single-day; exact day TBD)", "cfp_status": "published", "topics": [ "Discrete diffusion & flow models", "Optimal transport", "Optimal & stochastic control", "Schrödinger bridges", "Discrete diffusion & adjoint-based samplers", "MCMC & variational inference", "AI for science", "Multimodal generation", "Reward alignment", "Inverse problems", "Benchmarks & datasets", "Foundation models & large-scale training/inference", "Architectures for non-autoregressive generation", "Systems & hardware for diffusion models" ], "important_dates": [ { "label": "Submissions Open", "date": "2026-07-20" }, { "label": "Paper Submission Deadline", "date": "2026-08-29" }, { "label": "Notification", "date": "2026-09-28" }, { "label": "Camera-ready (tentative)", "date": "2026-10-20" }, { "label": "Workshop", "date": "2026-12-11 or 2026-12-12" } ], "organizers": [ "Minhyuk Sung (KAIST)", "Jaihoon Kim (KAIST)", "Molei Tao (Georgia Tech)", "Pranam Chatterjee (University of Pennsylvania)", "Sophia Tang (University of Pennsylvania)", "Nolan Dey (Cerebras Systems)", "Subham Sekhar Sahoo (MBZUAI)" ], "speakers": [ "Stefano Ermon (Stanford University)", "Pavlo Molchanov (NVIDIA Research)", "Rianne van den Berg (Microsoft Research)", "Sitan Chen (Harvard University)", "Chongxuan Li (Renmin University)", "Jiaxin Shi (Meta Superintelligence Labs)" ], "host_url": "https://bento-neurips.github.io/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/BeNTo", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/BeNTo", "location": "Sydney, Australia", "city": "Sydney", "workshop_date": "", "contact": "bento-neurips@googlegroups.com", "tracks": [ { "key": "BeNTo", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/BeNTo", "submission_dates_raw": "Submission Start: Aug 05 2026 11:59PM UTC-0, Submission Deadline: Aug 31 2026 11:59PM UTC-0" } ], "group": "fm", "group_label": "Foundation Models & LLMs", "corpus": "Beyond Next Token Prediction: Diffusion and Flow Models for Next-Generation Decoding BeNTo Workshop NeurIPS 2026 A NeurIPS 2026 workshop exploring the theory, algorithms, applications, and systems of discrete diffusion and flow models for parallel, non-causal generation — moving beyond fixed-order autoregressive next-token prediction — across two tracks on foundations and on applications/systems. Discrete diffusion & flow models Optimal transport Optimal & stochastic control Schrödinger bridges Discrete diffusion & adjoint-based samplers MCMC & variational inference AI for science Multimodal generation Reward alignment Inverse problems Benchmarks & datasets Foundation models & large-scale training/inference Architectures for non-autoregressive generation Systems & hardware for diffusion models Beyond Next-Token Prediction: Diffusion & Flow Models for Next-Generation Decoding\nBeyond next-token prediction — exploring the theory, algorithms, and applications of discrete diffusion and flow models for parallel, non-causal generation.\n\nOverview — Rethinking the order of generation.\nAutoregressive models produce their output one element at a time, in a fixed order. The recipe is remarkably general, and it has carried the field a long way. However, it also ties the order of computation to the order of the result, leaving parallelism, revision, and control hard to reach. Discrete diffusion and flow models take a different route. They generate by iterative denoising: refining many positions at once, and revisiting earlier choices as a sample takes shape.\n\nThat shift brings a different set of computational properties — non-causal, parallel, and controllable — along with open questions spanning mathematics, algorithms, and engineering, and a growing body of work that combines the two paradigms rather than choosing between them. This workshop joins two conversations: the theory and algorithms that make discrete generative models work, and the applications and systems that put them to use.\n\nTrack 01 — Depth: Theories & Algorithms\nDiscrete diffusion and flow models at the intersection of generative modeling, optimal transport, and stochastic optimal control.\n- Generative Modeling: Discrete diffusion & flow models, optimal transport, optimal control, Schrödinger bridges.\n- Probabilistic Inference: Discrete diffusion samplers, adjoint-based samplers, advanced MCMC, variational inference.\n\nTrack 02 — Breadth: Applications & Systems\nNovel applications and scalable systems that exploit non-causal, parallel generation.\n- Applications: AI for science, multimodal generation, reward alignment, inverse problems, benchmarks & datasets.\n- Systems & Empirical Analysis: Foundation models, large-scale training/inference, network architectures.\n\nScope — Topics of interest\nThis workshop considers, but is not limited to, the following topics.\n- Discrete diffusion & flow models\n- Optimal transport\n- Optimal & stochastic control\n- Schrödinger bridges\n- Discrete diffusion & adjoint-based samplers\n- MCMC & variational inference\n- AI for science\n- Multimodal generation\n- Reward alignment\n- Inverse problems\n- Benchmarks & datasets\n- Foundation models & large-scale training/inference\n- Architectures for non-autoregressive generation\n- Systems & hardware for diffusion models\n\nCall for Papers — Share your work.\nWhat to submit: We invite submissions on any topic within the workshop's scope, across both tracks, from theory and algorithms to applications and systems. All submissions are handled through OpenReview.\nPage limit: Submissions may be either 4 or 8 pages, excluding references and appendices.\nPeer review: Each submission receives 3 reviews; each reviewer handles at most 3 papers.\nMutual-review policy: At least one qualified author per submission serves as a reviewer.\nTravel grants for students and early-career researchers with accepted papers.\nPresentation: Accepted papers are presented as posters, with a selection of contributed orals (6 contributed orals, 10 min each). A Best Paper Award recognizes outstanding work.\nSubmissions are managed through OpenReview, at NeurIPS.cc/2026/Workshop/BeNTo. Questions? bento-neurips@googlegroups.com\n\nImportant Dates (all deadlines 23:59 AoE; dates indicative and subject to change)\n- Jul 20, 2026: Call for Papers & submissions open (OpenReview)\n- Aug 29, 2026: Paper submission deadline\n- Aug 30 – Sep 20, 2026: Peer review period\n- Sep 21 – 25, 2026: Meta-review & discussion\n- Sep 28, 2026: Accept / reject notification\n- Oct 20, 2026: Camera-ready deadline (tentative)\n- Dec 11 or 12, 2026: Workshop @ NeurIPS 2026 (single-day; exact day TBD)" }, { "key": "InfPriv", "title": "Beyond Private Training: The New Landscape of AI Privacy", "subtitle": "InfPriv 2026", "summary": "A NeurIPS 2026 workshop calling for a paradigm shift in AI privacy from the training phase toward inference-time and non-finetuning settings, covering agentic privacy, inference-only private synthetic data, in-context learning privacy, using AI for privacy, and new privacy benchmarks and evaluation.", "cfp_full": "About / Overview\n\nAs AI applications rapidly pivot to non-finetuning paradigms - such as autonomous agents, in-context learning, and inference-only synthetic data generation - we gather privacy theorists and applied practitioners to redirect attention to emerging post-training privacy paradigms.\n\nWhile the traditional setting (such as DP-SGD) disproportionately focuses on training-phase noise injection - which is computationally prohibitive for foundation models - the emerging frontier shifts attention to inference-time safety. This includes local privacy filters, agentic privacy guardrails, in-context leakage defense, and non-finetuning private synthesis. Under this new paradigm, we also explore AI for Privacy - leveraging LLMs as active infrastructure to enforce, redact, and verify mathematical guarantees (e.g., through Privacy Filters and benchmarks like DPrivBench).\n\nCall for Papers: Bridging the Academia-Industry Privacy Mismatch\n\nHistorically, the pursuit of privacy-preserving AI has been heavily anchored to the training phase, relying predominantly on training-time mechanisms. However, training or even fine-tuning state-of-the-art foundation models is computationally prohibitive for most practitioners. As a result, the landscape of AI applications has rapidly shifted toward efficient, inference-time, and non-finetuning paradigms, such as autonomous agents, in-context learning, and inference-only synthetic data generation. This workshop aims to call for a crucial paradigm shift, moving the privacy discourse beyond the training stage. We gather researchers from both industry and academia to collectively define the most pressing emerging privacy problems, synchronize theoretical rigor with production-level deployments, and explore how AI models can actively help us with privacy (AI for Privacy) - leveraging LLMs as active infrastructure to enforce, redact, and verify mathematical guarantees (e.g., through Privacy Filters and benchmarks like DPrivBench).\n\nWe welcome submissions of up to 4 pages presenting original research or work-in-progress on topics including:\n\n- Privacy for Agentic Systems: Understanding and mitigating privacy risks in single-agent and multi-agent systems, from identifying and measuring vulnerabilities to developing architectures that protect sensitive information.\n- Inference-only Private Synthetic Data Generation: Methodologies for generating differentially private synthetic data relying strictly on inference-time techniques without model training. Examples include private evolution, private prediction, and more.\n- Privacy in In-Context Learning: Risks of prompt leakage, PII extraction from context windows, and techniques for privacy-preserving in-context learning.\n- AI for Privacy: The application of AI models to automate data sanitization, construct privacy filters, optimize the design of private algorithms, and advance the theoretical foundations of Differential Privacy (DP).\n- Other Emerging Inference-Time AI Privacy Problems: Exploring new privacy problems and vulnerabilities that arise specifically at the inference-time or deployment-time AI scenarios.\n- Benchmarks and Evaluation: Designing new benchmarks, auditing frameworks, and metrics to systematically evaluate the privacy of AI systems and measure how effectively AI can enforce privacy.\n\nSubmission Instructions\n\nWe invite researchers to submit original work or extended abstracts aligning with the emerging frontiers of inference-time AI privacy and models utilized for safety. Short submissions (up to 4 pages) in NeurIPS format. Submissions must be written in English and formatted using the official NeurIPS template, using a strict double-blind setup. Ensure your submission is completely anonymized. Submissions should be in PDF format with an anonymous list of authors. Program committee members will review at most 3 submissions per submission.\n\nWe enforce a strict reference hallucination detection protocol on all submissions. Papers found to contain fabricated references, AI-generated citations, or significant unsubstantiated claims will be rejected automatically without review.\n\nThis is a full-day workshop, held in-person with live-stream. The workshop features invited lectures, oral presentation sessions, poster sessions, and an expert panel discussion.\n\nImportant Dates\n- Submission Deadline: September 07, 2026\n- Author Notification: September 29, 2026\n- Camera-ready Deadline: November 30, 2026\n- Workshop Date: December 11 or 12, 2026", "cfp_status": "published", "topics": [ "Privacy for Agentic Systems (privacy risks and protective architectures in single-agent and multi-agent systems)", "Inference-only Private Synthetic Data Generation (DP synthetic data via inference-time techniques such as private evolution and private prediction)", "Privacy in In-Context Learning (prompt leakage, PII extraction from context windows, privacy-preserving in-context learning)", "AI for Privacy (AI models to automate data sanitization, construct privacy filters, optimize private algorithms, advance DP foundations)", "Other Emerging Inference-Time AI Privacy Problems (new vulnerabilities at inference/deployment time)", "Benchmarks and Evaluation (new benchmarks, auditing frameworks, and metrics for AI privacy)" ], "important_dates": [ { "label": "Submission Start", "date": "2026-07-20" }, { "label": "Submission Deadline", "date": "2026-09-07" }, { "label": "Author Notification", "date": "2026-09-29" }, { "label": "Camera-ready Deadline", "date": "2026-11-30" }, { "label": "Workshop Date", "date": "December 11 or 12, 2026" } ], "organizers": [ "Eli Chien (National Taiwan University)", "Ruihan Wu (OpenAI)", "Kamalika Chaudhuri (Google DeepMind / UCSD)", "Yu-Xiang Wang (UCSD)", "Niloofar Mireshghallah (humans& / Carnegie Mellon University)", "Erchi Wang (UC San Diego)", "Jiachen \"Tianhao\" Wang (Princeton University)", "Antti Honkela (University of Helsinki)" ], "speakers": [ "Zinan Lin (Microsoft Research, Redmond)", "Chulin Xie (Google DeepMind, London)", "Mihaela van der Schaar (University of Cambridge)", "Dinh Thai Hoang (University of Technology Sydney)", "Saeed Mahloujifar (FAIR Labs, Meta)" ], "host_url": "https://beyond-private-training.ai.studio/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/InfPriv", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/InfPriv", "location": "Sydney", "city": "Sydney", "workshop_date": "", "contact": "neurips2026.infpriv@gmail.com", "tracks": [ { "key": "InfPriv", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/InfPriv", "submission_dates_raw": "Submission Start: Jul 20 2026 11:59AM UTC-0, Submission Deadline: Sep 08 2026 12:00PM UTC-0" } ], "group": "trust", "group_label": "Trustworthy & Safe AI", "corpus": "Beyond Private Training: The New Landscape of AI Privacy InfPriv 2026 A NeurIPS 2026 workshop calling for a paradigm shift in AI privacy from the training phase toward inference-time and non-finetuning settings, covering agentic privacy, inference-only private synthetic data, in-context learning privacy, using AI for privacy, and new privacy benchmarks and evaluation. Privacy for Agentic Systems (privacy risks and protective architectures in single-agent and multi-agent systems) Inference-only Private Synthetic Data Generation (DP synthetic data via inference-time techniques such as private evolution and private prediction) Privacy in In-Context Learning (prompt leakage, PII extraction from context windows, privacy-preserving in-context learning) AI for Privacy (AI models to automate data sanitization, construct privacy filters, optimize private algorithms, advance DP foundations) Other Emerging Inference-Time AI Privacy Problems (new vulnerabilities at inference/deployment time) Benchmarks and Evaluation (new benchmarks, auditing frameworks, and metrics for AI privacy) About / Overview\n\nAs AI applications rapidly pivot to non-finetuning paradigms - such as autonomous agents, in-context learning, and inference-only synthetic data generation - we gather privacy theorists and applied practitioners to redirect attention to emerging post-training privacy paradigms.\n\nWhile the traditional setting (such as DP-SGD) disproportionately focuses on training-phase noise injection - which is computationally prohibitive for foundation models - the emerging frontier shifts attention to inference-time safety. This includes local privacy filters, agentic privacy guardrails, in-context leakage defense, and non-finetuning private synthesis. Under this new paradigm, we also explore AI for Privacy - leveraging LLMs as active infrastructure to enforce, redact, and verify mathematical guarantees (e.g., through Privacy Filters and benchmarks like DPrivBench).\n\nCall for Papers: Bridging the Academia-Industry Privacy Mismatch\n\nHistorically, the pursuit of privacy-preserving AI has been heavily anchored to the training phase, relying predominantly on training-time mechanisms. However, training or even fine-tuning state-of-the-art foundation models is computationally prohibitive for most practitioners. As a result, the landscape of AI applications has rapidly shifted toward efficient, inference-time, and non-finetuning paradigms, such as autonomous agents, in-context learning, and inference-only synthetic data generation. This workshop aims to call for a crucial paradigm shift, moving the privacy discourse beyond the training stage. We gather researchers from both industry and academia to collectively define the most pressing emerging privacy problems, synchronize theoretical rigor with production-level deployments, and explore how AI models can actively help us with privacy (AI for Privacy) - leveraging LLMs as active infrastructure to enforce, redact, and verify mathematical guarantees (e.g., through Privacy Filters and benchmarks like DPrivBench).\n\nWe welcome submissions of up to 4 pages presenting original research or work-in-progress on topics including:\n\n- Privacy for Agentic Systems: Understanding and mitigating privacy risks in single-agent and multi-agent systems, from identifying and measuring vulnerabilities to developing architectures that protect sensitive information.\n- Inference-only Private Synthetic Data Generation: Methodologies for generating differentially private synthetic data relying strictly on inference-time techniques without model training. Examples include private evolution, private prediction, and more.\n- Privacy in In-Context Learning: Risks of prompt leakage, PII extraction from context windows, and techniques for privacy-preserving in-context learning.\n- AI for Privacy: The application of AI models to automate data sanitization, construct privacy filters, optimize the design of private algorithms, and advance the theoretical foundations of Differential Privacy (DP).\n- Other Emerging Inference-Time AI Privacy Problems: Exploring new privacy problems and vulnerabilities that arise specifically at the inference-time or deployment-time AI scenarios.\n- Benchmarks and Evaluation: Designing new benchmarks, auditing frameworks, and metrics to systematically evaluate the privacy of AI systems and measure how effectively AI can enforce privacy.\n\nSubmission Instructions\n\nWe invite researchers to submit original work or extended abstracts aligning with the emerging frontiers of inference-time AI privacy and models utilized for safety. Short submissions (up to 4 pages) in NeurIPS format. Submissions must be written in English and formatted using the official NeurIPS template, using a strict double-blind setup. Ensure your submission is completely anonymized. Submissions should be in PDF format with an anonymous list of authors. Program committee members will review at most 3 submissions per submission.\n\nWe enforce a strict reference hallucination detection protocol on all submissions. Papers found to contain fabricated references, AI-generated citations, or significant unsubstantiated claims will be rejected automatically without review.\n\nThis is a full-day workshop, held in-person with live-stream. The workshop features invited lectures, oral presentation sessions, poster sessions, and an expert panel discussion.\n\nImportant Dates\n- Submission Deadline: September 07, 2026\n- Author Notification: September 29, 2026\n- Camera-ready Deadline: November 30, 2026\n- Workshop Date: December 11 or 12, 2026" }, { "key": "DevAI", "title": "DevAI Workshop: Developmental Perspectives on AI", "subtitle": "DevAI 2026", "summary": "A workshop bringing together researchers from AI, cognitive science, neuroscience, and developmental psychology to explore what human development can teach us about building more robust, efficient, and human-like artificial intelligence.", "cfp_full": "Developmental Perspectives on AI at NeurIPS'26\nHow Does Intelligence Emerge?\nDevAI brings together researchers from AI, cognitive science, neuroscience, and developmental psychology to explore what human development can teach us about building more robust, efficient, and human-like artificial intelligence.\n\nCall for Papers\nWe encourage the submission of research exploring developmental perspectives on AI, spanning developmental psychology, neuroscience, cognitive science, and machine learning, as well as papers introducing new developmental datasets and benchmarks for AI.\n\nTopics of Interest\n- Core Knowledge and Inductive Biases\n- Development of Representations and Concepts\n- Learning and Generalization from Limited Experience\n- Development of Multimodal Representations\n- Social Learning and Social Cognition in Humans and Machines\n- Intuitive Physics, Causal Reasoning, and World Models\n- Few-Shot and Data-Efficient Learning Inspired by Development\n- Development of Representations in Brains and Machines\n- Visual Representation Learning Inspired by Development\n- Language Acquisition and Developmental Language Learning\n- Developmental Neuroscience and Computational Accounts of Early Learning\n- Developmental Datasets and Naturalistic Learning Environments\n- Computational Models of Cognitive Development\n- Developmentally Inspired Machine Learning Architectures\n- Developmental Benchmarks for AI Systems\n- AI as a Tool for Studying Cognitive and Brain Development\n\nImportant Dates\nRegular Papers (Archival Track):\n- Paper submission: August 29th 2026 (23:59 AoE)\n- Author Notification: September 29th 2026\n- Camera ready paper submission: October 8th 2026\nExtended Abstracts (Non-Archival Track):\n- Paper submissions: October 13th 2026 (23:59 AoE)\n- Author Notification: October 20th 2026", "cfp_status": "published", "topics": [ "Core Knowledge and Inductive Biases", "Development of Representations and Concepts", "Learning and Generalization from Limited Experience", "Development of Multimodal Representations", "Social Learning and Social Cognition in Humans and Machines", "Intuitive Physics, Causal Reasoning, and World Models", "Few-Shot and Data-Efficient Learning Inspired by Development", "Development of Representations in Brains and Machines", "Visual Representation Learning Inspired by Development", "Language Acquisition and Developmental Language Learning", "Developmental Neuroscience and Computational Accounts of Early Learning", "Developmental Datasets and Naturalistic Learning Environments", "Computational Models of Cognitive Development", "Developmentally Inspired Machine Learning Architectures", "Developmental Benchmarks for AI Systems", "AI as a Tool for Studying Cognitive and Brain Development" ], "important_dates": [ { "label": "Regular Papers (Archival) - Paper submission", "date": "2026-08-29" }, { "label": "Regular Papers (Archival) - Author Notification", "date": "2026-09-29" }, { "label": "Regular Papers (Archival) - Camera ready", "date": "2026-10-08" }, { "label": "Extended Abstracts (Non-Archival) - Paper submission", "date": "2026-10-13" }, { "label": "Extended Abstracts (Non-Archival) - Author Notification", "date": "2026-10-20" }, { "label": "Workshop", "date": "2026-12-12" } ], "organizers": [ "Shify Treger (Weizmann Institute of Science)", "Yehonatan Avidan (Hebrew University)", "Mengmi Zhang (Nanyang Technological University)", "Kelsey Allen (University of British Columbia)", "Shimon Ullman (Weizmann Institute of Science)" ], "speakers": [ "Joshua Tenenbaum (MIT)", "Cameron Ellis (Stanford)", "Ashley Thomas (Harvard)", "Brenden Lake (Princeton)", "Daniel Zoran (DeepMind, TBC)", "Uri Hasson (Princeton, TBC)", "Elizabeth Spelke (Harvard)", "Bria Long (UCSD)" ], "host_url": "https://sites.google.com/view/devai-workshop-2026/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/DevAI", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/DevAI", "location": "Atlanta, USA", "city": "Atlanta", "workshop_date": "2026-12-09", "contact": "shify.treger@weizmann.ac.il", "tracks": [ { "key": "DevAI", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/DevAI", "submission_dates_raw": "" } ], "group": "society", "group_label": "Society, Applications & Community", "corpus": "DevAI Workshop: Developmental Perspectives on AI DevAI 2026 A workshop bringing together researchers from AI, cognitive science, neuroscience, and developmental psychology to explore what human development can teach us about building more robust, efficient, and human-like artificial intelligence. Core Knowledge and Inductive Biases Development of Representations and Concepts Learning and Generalization from Limited Experience Development of Multimodal Representations Social Learning and Social Cognition in Humans and Machines Intuitive Physics, Causal Reasoning, and World Models Few-Shot and Data-Efficient Learning Inspired by Development Development of Representations in Brains and Machines Visual Representation Learning Inspired by Development Language Acquisition and Developmental Language Learning Developmental Neuroscience and Computational Accounts of Early Learning Developmental Datasets and Naturalistic Learning Environments Computational Models of Cognitive Development Developmentally Inspired Machine Learning Architectures Developmental Benchmarks for AI Systems AI as a Tool for Studying Cognitive and Brain Development Developmental Perspectives on AI at NeurIPS'26\nHow Does Intelligence Emerge?\nDevAI brings together researchers from AI, cognitive science, neuroscience, and developmental psychology to explore what human development can teach us about building more robust, efficient, and human-like artificial intelligence.\n\nCall for Papers\nWe encourage the submission of research exploring developmental perspectives on AI, spanning developmental psychology, neuroscience, cognitive science, and machine learning, as well as papers introducing new developmental datasets and benchmarks for AI.\n\nTopics of Interest\n- Core Knowledge and Inductive Biases\n- Development of Representations and Concepts\n- Learning and Generalization from Limited Experience\n- Development of Multimodal Representations\n- Social Learning and Social Cognition in Humans and Machines\n- Intuitive Physics, Causal Reasoning, and World Models\n- Few-Shot and Data-Efficient Learning Inspired by Development\n- Development of Representations in Brains and Machines\n- Visual Representation Learning Inspired by Development\n- Language Acquisition and Developmental Language Learning\n- Developmental Neuroscience and Computational Accounts of Early Learning\n- Developmental Datasets and Naturalistic Learning Environments\n- Computational Models of Cognitive Development\n- Developmentally Inspired Machine Learning Architectures\n- Developmental Benchmarks for AI Systems\n- AI as a Tool for Studying Cognitive and Brain Development\n\nImportant Dates\nRegular Papers (Archival Track):\n- Paper submission: August 29th 2026 (23:59 AoE)\n- Author Notification: September 29th 2026\n- Camera ready paper submission: October 8th 2026\nExtended Abstracts (Non-Archival Track):\n- Paper submissions: October 13th 2026 (23:59 AoE)\n- Author Notification: October 20th 2026" }, { "key": "DiffuLM", "title": "Diffusion Language Models: Foundations, Efficiency, and Reasoning (DiffuLM)", "subtitle": "The first NeurIPS workshop dedicated to diffusion language models", "summary": "DiffuLM @ NeurIPS 2026 is the first NeurIPS workshop dedicated to diffusion language models, spanning their foundations (theory & scaling), efficiency (parallel decoding, fast sampling, serving), and reasoning (self-correction, test-time scaling, RL post-training).", "cfp_full": "DiffuLM @ NeurIPS 2026 - Diffusion Language Models: Foundations, Efficiency, and Reasoning\nThe first NeurIPS workshop dedicated to diffusion language models.\n\nWorkshop focus - From foundations to intelligence\n- Foundations - Theory & Scaling: Unifying diffusion formulations, scaling laws, training objectives, probabilistic inference, and model architectures.\n- Performance - Efficiency: Parallel decoding, fast sampling, long-context generation, serving, evaluation, and hardware-aware systems.\n- Intelligence - Reasoning: Global self-correction, diverse solution paths, test-time scaling, planning, code generation, and RL post-training.\n\nCall for Papers\n\nWe invite extended abstracts and short papers advancing diffusion language models and related non-autoregressive methods across foundations, efficiency, reasoning, and systems.\n\nWhat we are looking for\n\nWe welcome theoretical, algorithmic, empirical, and systems work that helps establish diffusion language models as a rigorous and practical alternative to autoregressive generation.\n\nSubmissions are non-archival and reviewed double-blind on OpenReview. Work already published at NeurIPS 2026 or another archival venue is not eligible. Every submission will receive at least three reviews. The workshop will recognize a Best Paper and a Best Student Paper Award.\n\nReview & policies (Shared requirements)\n- Submission and review through OpenReview\n- Double-blind review\n- At least three reviews per submission\n- Non-archival; previously published archival work is not eligible\n- Explicit COI declarations on OpenReview\n- Organizer recusal for own-institution submissions and active collaborators\n\nSubmission tracks\n- Track 01 - Extended Abstracts: Up to 4 pages of main text. Work-in-progress, preliminary results, and position papers.\n- Track 02 - Short Papers: Up to 8 pages of main text. More complete contributions with experiments or theoretical results.\n\nFormatting requirement - NeurIPS 2026 style: Papers in both tracks must be formatted with the NeurIPS 2026 style template, available from the NeurIPS 2026 Call for Papers page.\n\nTopics of Interest\n- Foundations: Categorical, masked, and discrete diffusion; Training objectives and probabilistic inference; Scaling laws and model architectures; Hybrid autoregressive-diffusion methods\n- Efficiency: Sampling and parallel decoding; Accelerated and adaptive inference; Long-context generation and serving; Evaluation and hardware-aware systems\n- Reasoning: Reasoning, planning, and code generation; Test-time scaling and self-correction; Controllable generation and post-training; Reinforcement learning and alignment\n\nKey Dates\n- Submission: August 29, 2026 (Anywhere on Earth)\n- Notification: September 29, 2026 (AoE)\n- Camera-ready: October 2026\n- Workshop: December 12, 2026 (Sydney)", "cfp_status": "published", "topics": [ "Categorical, masked, and discrete diffusion", "Training objectives and probabilistic inference", "Scaling laws and model architectures", "Hybrid autoregressive-diffusion methods", "Sampling and parallel decoding", "Accelerated and adaptive inference", "Long-context generation and serving", "Evaluation and hardware-aware systems", "Reasoning, planning, and code generation", "Test-time scaling and self-correction", "Controllable generation and post-training", "Reinforcement learning and alignment" ], "important_dates": [ { "label": "Submission deadline (AoE)", "date": "2026-08-29" }, { "label": "Notification (AoE)", "date": "2026-09-29" }, { "label": "Camera-ready", "date": "October 2026" }, { "label": "Workshop", "date": "2026-12-12" } ], "organizers": [ "Jinjie Ni (Google DeepMind)", "Shansan Gong (HKU)", "Amin Karimi Monsefi (Ohio State University)", "Yuchen Zhu (Georgia Tech)", "Irina Belousova (Apple)", "Michael Qizhe Shieh (NUS)", "Yizhe Zhang (Meta)", "Pavlo Molchanov (NVIDIA Research)" ], "speakers": [ "Stefano Ermon (Stanford University)", "Subham Sahoo (MBZUAI)", "Arash Vahdat (NVIDIA Research)", "Volodymyr Kuleshov (Cornell / Inception Labs)", "Itai Gat (Meta Superintelligence Lab)", "Aditya Grover (UCLA / Inception)" ], "host_url": "https://7amin.github.io/diffulm-neurips2026/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/DiffuLM", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/DiffuLM", "location": "Sydney, Australia", "city": "Sydney", "workshop_date": "2026-12-10", "contact": "amin.karimi1992@gmail.com", "tracks": [ { "key": "DiffuLM", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/DiffuLM", "submission_dates_raw": "" } ], "group": "fm", "group_label": "Foundation Models & LLMs", "corpus": "Diffusion Language Models: Foundations, Efficiency, and Reasoning (DiffuLM) The first NeurIPS workshop dedicated to diffusion language models DiffuLM @ NeurIPS 2026 is the first NeurIPS workshop dedicated to diffusion language models, spanning their foundations (theory & scaling), efficiency (parallel decoding, fast sampling, serving), and reasoning (self-correction, test-time scaling, RL post-training). Categorical, masked, and discrete diffusion Training objectives and probabilistic inference Scaling laws and model architectures Hybrid autoregressive-diffusion methods Sampling and parallel decoding Accelerated and adaptive inference Long-context generation and serving Evaluation and hardware-aware systems Reasoning, planning, and code generation Test-time scaling and self-correction Controllable generation and post-training Reinforcement learning and alignment DiffuLM @ NeurIPS 2026 - Diffusion Language Models: Foundations, Efficiency, and Reasoning\nThe first NeurIPS workshop dedicated to diffusion language models.\n\nWorkshop focus - From foundations to intelligence\n- Foundations - Theory & Scaling: Unifying diffusion formulations, scaling laws, training objectives, probabilistic inference, and model architectures.\n- Performance - Efficiency: Parallel decoding, fast sampling, long-context generation, serving, evaluation, and hardware-aware systems.\n- Intelligence - Reasoning: Global self-correction, diverse solution paths, test-time scaling, planning, code generation, and RL post-training.\n\nCall for Papers\n\nWe invite extended abstracts and short papers advancing diffusion language models and related non-autoregressive methods across foundations, efficiency, reasoning, and systems.\n\nWhat we are looking for\n\nWe welcome theoretical, algorithmic, empirical, and systems work that helps establish diffusion language models as a rigorous and practical alternative to autoregressive generation.\n\nSubmissions are non-archival and reviewed double-blind on OpenReview. Work already published at NeurIPS 2026 or another archival venue is not eligible. Every submission will receive at least three reviews. The workshop will recognize a Best Paper and a Best Student Paper Award.\n\nReview & policies (Shared requirements)\n- Submission and review through OpenReview\n- Double-blind review\n- At least three reviews per submission\n- Non-archival; previously published archival work is not eligible\n- Explicit COI declarations on OpenReview\n- Organizer recusal for own-institution submissions and active collaborators\n\nSubmission tracks\n- Track 01 - Extended Abstracts: Up to 4 pages of main text. Work-in-progress, preliminary results, and position papers.\n- Track 02 - Short Papers: Up to 8 pages of main text. More complete contributions with experiments or theoretical results.\n\nFormatting requirement - NeurIPS 2026 style: Papers in both tracks must be formatted with the NeurIPS 2026 style template, available from the NeurIPS 2026 Call for Papers page.\n\nTopics of Interest\n- Foundations: Categorical, masked, and discrete diffusion; Training objectives and probabilistic inference; Scaling laws and model architectures; Hybrid autoregressive-diffusion methods\n- Efficiency: Sampling and parallel decoding; Accelerated and adaptive inference; Long-context generation and serving; Evaluation and hardware-aware systems\n- Reasoning: Reasoning, planning, and code generation; Test-time scaling and self-correction; Controllable generation and post-training; Reinforcement learning and alignment\n\nKey Dates\n- Submission: August 29, 2026 (Anywhere on Earth)\n- Notification: September 29, 2026 (AoE)\n- Camera-ready: October 2026\n- Workshop: December 12, 2026 (Sydney)" }, { "key": "DynaFront", "title": "Dynamics at the Frontiers of Optimization, Sampling, and Games", "subtitle": "DynaFront@NeurIPS26", "summary": "DynaFront highlights the unifying role of dynamical systems across optimization, sampling, and game theory, convening experts to foster cross-disciplinary dialogue and lower the barrier to entry for these foundational methods, with an emphasis on emerging machine learning applications such as diffusion models, distributed and adversarial training, and agentic AI.", "cfp_full": "DynaFront 2: Dynamics at the Frontiers of Optimization, Sampling, and Games, 2nd Edition\nWorkshop at NeurIPS 2026\nSaturday/Sunday, December 12/13, 2026\nAtlanta, GA\n\nDynamical systems have played an important role in the analysis and design of algorithms. Ideas ranging from variational methods, differential and symplectic geometry, numerical analysis, and control theory have paved the way for establishing non-asymptotic convergence guarantees in optimization, sampling, and equilibrium computation in games. Yet, the distinct mathematical backbone of these tools often creates barriers to entry for researchers and practitioners in machine learning.\nThis workshop aims to lower that barrier by highlighting the unifying role of dynamical systems across these domains. We will convene optimization, sampling, and game theory experts to foster cross-disciplinary dialogue and collaboration. Emphasis will be placed on emerging applications in machine learning, such as diffusion models, distributed and adversarial training, and agentic AI, where dynamical systems perspectives are increasingly central. Through a combination of talks, posters, and open discussions, we hope to catalyze new collaborations and broaden the accessibility of these foundational methods.\nThis is the second edition of the DynaFront workshop; the first one was held during NeurIPS 2025 in San Diego, CA.\nThis workshop is held during NeurIPS 2026 in Atlanta, GA.\n\nSpeakers:\nProf. Niao He, ETH Zurich\nProf. Jianfeng Lu, Duke University\nProf. Santosh Vempala, Georgia Institute of Technology\nProf. Lenka Zdeborova, EPFL\nProf. Tong Zhang, University of Illinois\nProf. Xunyu Zhou, Columbia University\n\nSubmission format: Papers should be 4 to 5 pages (excluding references and supplementary materials), prepared using the workshop template. Submissions are handled through OpenReview. Reviewing is a single round with no author response, and at least one author from each submission should volunteer as a reviewer.\n\nImportant Dates:\nSubmission Deadline: August 29 / September 4, 2026 (AoE)\nAcceptance Notification: September 29, 2026 (AoE)\nCamera Ready: TBA\nWorkshop: December 12/13, 2026\nLocation: Atlanta, GA\n\nImportant Links:\nSubmission portal: https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/DynaFront\nReviewer application form: https://forms.gle/vVn5cptJB6ggT37n6", "cfp_status": "published", "topics": [ "Optimization", "Sampling", "Equilibrium computation in games", "Dynamical systems perspectives on algorithms", "Diffusion models", "Distributed and adversarial training", "Agentic AI", "Langevin-based sampling and constrained distributions", "Nash equilibrium computation" ], "important_dates": [ { "label": "Submission Deadline", "date": "2026-09-04" }, { "label": "Acceptance Notification", "date": "2026-09-29" }, { "label": "Camera Ready", "date": "TBA" }, { "label": "Workshop", "date": "December 12/13, 2026" } ], "organizers": [ "Zhiyu He (ETHZ / Max Planck Institute for Intelligent Systems, Tübingen)", "Michael I. Jordan (Inria Paris and University of California, Berkeley)", "Jiaming Liang (University of Rochester)", "Wenlong Mou (University of Toronto)", "Michael Mühlebach (Max Planck Institute for Intelligent Systems, Tübingen)", "Purnamrita Sarkar (University of Texas)", "Molei Tao (Georgia Institute of Technology)", "Andre Wibisono (Yale University)" ], "speakers": [ "Niao He (ETH Zurich)", "Jianfeng Lu (Duke University)", "Santosh Vempala (Georgia Institute of Technology)", "Lenka Zdeborova (EPFL)", "Tong Zhang (University of Illinois)", "Xunyu Zhou (Columbia University)" ], "host_url": "https://sites.google.com/view/dynafrontneurips26", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/DynaFront", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/DynaFront", "location": "Atlanta, USA", "city": "Atlanta", "workshop_date": "2026-12-12", "contact": "dynafront.neurips@gmail.com", "tracks": [ { "key": "DynaFront", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/DynaFront", "submission_dates_raw": "" } ], "group": "theory", "group_label": "Theory & Methods", "corpus": "Dynamics at the Frontiers of Optimization, Sampling, and Games DynaFront@NeurIPS26 DynaFront highlights the unifying role of dynamical systems across optimization, sampling, and game theory, convening experts to foster cross-disciplinary dialogue and lower the barrier to entry for these foundational methods, with an emphasis on emerging machine learning applications such as diffusion models, distributed and adversarial training, and agentic AI. Optimization Sampling Equilibrium computation in games Dynamical systems perspectives on algorithms Diffusion models Distributed and adversarial training Agentic AI Langevin-based sampling and constrained distributions Nash equilibrium computation DynaFront 2: Dynamics at the Frontiers of Optimization, Sampling, and Games, 2nd Edition\nWorkshop at NeurIPS 2026\nSaturday/Sunday, December 12/13, 2026\nAtlanta, GA\n\nDynamical systems have played an important role in the analysis and design of algorithms. Ideas ranging from variational methods, differential and symplectic geometry, numerical analysis, and control theory have paved the way for establishing non-asymptotic convergence guarantees in optimization, sampling, and equilibrium computation in games. Yet, the distinct mathematical backbone of these tools often creates barriers to entry for researchers and practitioners in machine learning.\nThis workshop aims to lower that barrier by highlighting the unifying role of dynamical systems across these domains. We will convene optimization, sampling, and game theory experts to foster cross-disciplinary dialogue and collaboration. Emphasis will be placed on emerging applications in machine learning, such as diffusion models, distributed and adversarial training, and agentic AI, where dynamical systems perspectives are increasingly central. Through a combination of talks, posters, and open discussions, we hope to catalyze new collaborations and broaden the accessibility of these foundational methods.\nThis is the second edition of the DynaFront workshop; the first one was held during NeurIPS 2025 in San Diego, CA.\nThis workshop is held during NeurIPS 2026 in Atlanta, GA.\n\nSpeakers:\nProf. Niao He, ETH Zurich\nProf. Jianfeng Lu, Duke University\nProf. Santosh Vempala, Georgia Institute of Technology\nProf. Lenka Zdeborova, EPFL\nProf. Tong Zhang, University of Illinois\nProf. Xunyu Zhou, Columbia University\n\nSubmission format: Papers should be 4 to 5 pages (excluding references and supplementary materials), prepared using the workshop template. Submissions are handled through OpenReview. Reviewing is a single round with no author response, and at least one author from each submission should volunteer as a reviewer.\n\nImportant Dates:\nSubmission Deadline: August 29 / September 4, 2026 (AoE)\nAcceptance Notification: September 29, 2026 (AoE)\nCamera Ready: TBA\nWorkshop: December 12/13, 2026\nLocation: Atlanta, GA\n\nImportant Links:\nSubmission portal: https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/DynaFront\nReviewer application form: https://forms.gle/vVn5cptJB6ggT37n6" }, { "key": "Meta-Agents", "title": "First Workshop on Meta Agents: Managing Agents that Manage Agents", "subtitle": "NeurIPS 2026 Workshop Meta-Agents", "summary": "A workshop on the responsible use of meta-agents: higher-order agents that build, optimize, and supervise other agents, covering the full lifecycle of design, training, evaluation, deployment, and oversight.", "cfp_full": "Managing Agents that Manage Agents\nWorkshop on Responsible Use of Meta-Agents\nNeurIPS 2026 · December 11-12, 2026 · Sydney, Australia\n\nMeta-agents are increasingly used to train, manage, and supervise other agents, and even people. How do we ensure this capability develops responsibly?\n\nAbout The Workshop\nMost agentic systems today rely on manual design for their training, harnesses, and goals. As models grow more capable, new systems are emerging to automate this. AlphaEvolve writes and tests its own programs; GEPA and Meta-Harness use agents to optimize the trajectories of other agents, beating reinforcement learning at a fraction of the cost. We call these higher-order agents that build, optimize, and supervise other agents meta-agents.\n\nMeta-agents will likely play an increasingly central role in how agentic systems are used and developed. Yet discourse on them stays dispersed across separate communities, from meta-learning to reinforcement learning to systems research. A systematic, interdisciplinary view is needed to address their potential for both use and misuse, convening core figures from these subfields alongside researchers in AI and organizational science.\n\nThe shift from manual to automated agent design raises open technical problems. A meta-agent can design the harness, train the agent system, and enable continual learning and self-improvement. When does a discovered or self-improved design actually generalize, and which optimization methods and learning signals improve a meta-agent's capabilities?\n\nAs capabilities scale, societal impact needs careful oversight. A meta-agent acts like a manager, decomposing objectives and assigning tasks to worker agents, so a misaligned one can spawn and coordinate swarms of subordinate agents toward the wrong goal. Meta-agents can also manage humans, with risks of economic disruption and eroded autonomy. So every result raises the same question: who evaluates and oversees an agent that builds or improves another agent, or itself? The workshop covers the full lifecycle, design, training, evaluation, deployment, and oversight, and every submission carries a responsible-use statement.\n\nTopics\nWe invite work across the topics below, capability first, then evaluation, oversight, and governance. Topics include but are not limited to:\n\n(1) Automated Design and Discovery of Agent Harnesses: Most agent harnesses are still built by hand. We invite work on searching the space of agent architectures, tools, and harnesses (Meta Agent Search, automated harness design), and on telling when a discovered design generalizes past the tasks it was tuned on.\n\n(2) Optimization and Post-Training of Compound Agentic Systems: Once an agent is more than one model call, the problem is optimizing the whole program. We seek work on optimizers for multi-module systems (DSPy, GEPA, Trace), credit assignment across modules, post-training of compound systems, inference-time search, and context optimization.\n\n(3) Self-Improvement and Open-Ended Evolution: A meta-agent can point at itself. We welcome work on agents that rewrite their own code (the Darwin Godel Machine, self-taught optimizers), self-play and co-evolution, open-endedness, and honest analyses of how self-improving agents drift or game their own reward.\n\n(4) Evaluation and Benchmarks for Meta-Agents: Does a discovered or self-improved agent actually generalize? We want benchmarks and tests for meta-level methods, the ones that answer this question rather than leaderboards that hide the failure cases.\n\n(5) Misalignment and Safety for Meta-Agents: Making agent execution observable and auditable, red-teaming systems that act on other agents, and detecting reward gaming, over-optimization, and drift before a system acts on other agents at scale.\n\n(6) Governance and Human Oversight of Meta-Agents: Rollback and halt controls a deployer can actually trigger, who signs off before a discovered agent ships, and how to govern agents that manage other agents, including perspectives from management science and other fields outside AI.\n\nCall For Papers\nManaging Agents that Manage Agents (NeurIPS 2026) invites submissions on architectures, algorithms, theory, empirical studies, benchmarks, and position papers about agents that design, optimize, supervise, train, or improve other agents. Submissions must present original, unpublished work that has not appeared at NeurIPS or other archival machine-learning venues.\n\nKey Dates\nSubmission Deadline: August 29, 2026, AoE\nNotification: on or before September 29, 2026, AoE\nWorkshop Date: December 11-12, 2026 (Sydney)\nAll deadlines follow the Anywhere on Earth (AoE) timezone.\n\nSubmission Site\nSubmissions are managed via OpenReview and remain private during review. All authors should maintain up-to-date OpenReview profiles for conflict-of-interest management and paper matching. Submit your paper at the OpenReview submission portal.\n\nScope\nWe welcome contributions across the topics above. Accepted papers are presented as posters, with a subset selected for oral or spotlight talks, and we give a Best Paper award and a Best Social Impact Paper award for the work that most thoughtfully addresses the societal implications of meta-agents. The workshop is in person at NeurIPS 2026 in Sydney.\n\nSubmission Guidelines\nFormatting: Submissions must be in English and use the NeurIPS 2026 workshop LaTeX template. Papers are submitted as a single PDF:\n- Full Papers: at most 9 pages (main text)\n- Short Papers: at most 4 pages (main text)\n- Demo Track: live demonstrations of meta-agent systems and tools, presented alongside the poster sessions.\n- Position Papers: on the governance, oversight, and societal impact of meta-agents.\nReferences and appendices do not count toward the page limit, but the main text must be self-contained.\n\nResponsible-Use Statement: Every submission also includes a short responsible-use statement covering the potential societal impacts of the proposed work and suggested mitigations. It is reviewed with the paper, and a missing one is grounds for desk rejection.\n\nAnonymity: The workshop uses double-blind review. Submissions must be anonymized, with author names, affiliations, and acknowledgments removed and prior work cited in the third person.\n\nNon-Archival Policy: The workshop is non-archival. Papers under review elsewhere are welcome, and accepted papers may be published at other venues afterward. Work already published at NeurIPS or other archival ML venues should not be submitted.\n\nContact: Email meta-agents-workshop@googlegroups.com.", "cfp_status": "published", "topics": [ "Automated Design and Discovery of Agent Harnesses", "Optimization and Post-Training of Compound Agentic Systems", "Self-Improvement and Open-Ended Evolution", "Evaluation and Benchmarks for Meta-Agents", "Misalignment and Safety for Meta-Agents", "Governance and Human Oversight of Meta-Agents" ], "important_dates": [ { "label": "Submission Deadline", "date": "2026-08-29" }, { "label": "Notification", "date": "2026-09-29" }, { "label": "Workshop Date", "date": "2026-12-11 to 2026-12-12" } ], "organizers": [ "Simon Yu (Northeastern University)", "Dilara Soylu (Stanford University)", "Derek Chong (Stanford University)", "Ananjan Nandi (Stanford University)", "Apurva Gandhi (Carnegie Mellon University)", "Jiuding Sun (Stanford University)", "Zichen Liu (Google DeepMind)", "Christopher D. Manning (Stanford University)", "Weiyan Shi (Northeastern University)" ], "speakers": [ "Graham Neubig (Carnegie Mellon University)", "Chelsea Finn (Stanford University)", "Jenny Zhang (Recursive Superintelligence & UBC)", "Robert Lange (Sakana AI)", "Bo Li (University of Illinois Urbana-Champaign)", "Hancheng Cao (Emory University)", "Yuandong Tian (Recursive)", "Chen Sun (Google DeepMind)" ], "host_url": "https://meta-agents-workshop.github.io/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/Meta-Agents", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/Meta-Agents", "location": "Sydney, Australia", "city": "Sydney", "workshop_date": "2026-12-11", "contact": "meta-agents-workshop@googlegroups.com", "tracks": [ { "key": "Meta-Agents", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/Meta-Agents", "submission_dates_raw": "" } ], "group": "agents", "group_label": "Agents & Agentic Systems", "corpus": "First Workshop on Meta Agents: Managing Agents that Manage Agents NeurIPS 2026 Workshop Meta-Agents A workshop on the responsible use of meta-agents: higher-order agents that build, optimize, and supervise other agents, covering the full lifecycle of design, training, evaluation, deployment, and oversight. Automated Design and Discovery of Agent Harnesses Optimization and Post-Training of Compound Agentic Systems Self-Improvement and Open-Ended Evolution Evaluation and Benchmarks for Meta-Agents Misalignment and Safety for Meta-Agents Governance and Human Oversight of Meta-Agents Managing Agents that Manage Agents\nWorkshop on Responsible Use of Meta-Agents\nNeurIPS 2026 · December 11-12, 2026 · Sydney, Australia\n\nMeta-agents are increasingly used to train, manage, and supervise other agents, and even people. How do we ensure this capability develops responsibly?\n\nAbout The Workshop\nMost agentic systems today rely on manual design for their training, harnesses, and goals. As models grow more capable, new systems are emerging to automate this. AlphaEvolve writes and tests its own programs; GEPA and Meta-Harness use agents to optimize the trajectories of other agents, beating reinforcement learning at a fraction of the cost. We call these higher-order agents that build, optimize, and supervise other agents meta-agents.\n\nMeta-agents will likely play an increasingly central role in how agentic systems are used and developed. Yet discourse on them stays dispersed across separate communities, from meta-learning to reinforcement learning to systems research. A systematic, interdisciplinary view is needed to address their potential for both use and misuse, convening core figures from these subfields alongside researchers in AI and organizational science.\n\nThe shift from manual to automated agent design raises open technical problems. A meta-agent can design the harness, train the agent system, and enable continual learning and self-improvement. When does a discovered or self-improved design actually generalize, and which optimization methods and learning signals improve a meta-agent's capabilities?\n\nAs capabilities scale, societal impact needs careful oversight. A meta-agent acts like a manager, decomposing objectives and assigning tasks to worker agents, so a misaligned one can spawn and coordinate swarms of subordinate agents toward the wrong goal. Meta-agents can also manage humans, with risks of economic disruption and eroded autonomy. So every result raises the same question: who evaluates and oversees an agent that builds or improves another agent, or itself? The workshop covers the full lifecycle, design, training, evaluation, deployment, and oversight, and every submission carries a responsible-use statement.\n\nTopics\nWe invite work across the topics below, capability first, then evaluation, oversight, and governance. Topics include but are not limited to:\n\n(1) Automated Design and Discovery of Agent Harnesses: Most agent harnesses are still built by hand. We invite work on searching the space of agent architectures, tools, and harnesses (Meta Agent Search, automated harness design), and on telling when a discovered design generalizes past the tasks it was tuned on.\n\n(2) Optimization and Post-Training of Compound Agentic Systems: Once an agent is more than one model call, the problem is optimizing the whole program. We seek work on optimizers for multi-module systems (DSPy, GEPA, Trace), credit assignment across modules, post-training of compound systems, inference-time search, and context optimization.\n\n(3) Self-Improvement and Open-Ended Evolution: A meta-agent can point at itself. We welcome work on agents that rewrite their own code (the Darwin Godel Machine, self-taught optimizers), self-play and co-evolution, open-endedness, and honest analyses of how self-improving agents drift or game their own reward.\n\n(4) Evaluation and Benchmarks for Meta-Agents: Does a discovered or self-improved agent actually generalize? We want benchmarks and tests for meta-level methods, the ones that answer this question rather than leaderboards that hide the failure cases.\n\n(5) Misalignment and Safety for Meta-Agents: Making agent execution observable and auditable, red-teaming systems that act on other agents, and detecting reward gaming, over-optimization, and drift before a system acts on other agents at scale.\n\n(6) Governance and Human Oversight of Meta-Agents: Rollback and halt controls a deployer can actually trigger, who signs off before a discovered agent ships, and how to govern agents that manage other agents, including perspectives from management science and other fields outside AI.\n\nCall For Papers\nManaging Agents that Manage Agents (NeurIPS 2026) invites submissions on architectures, algorithms, theory, empirical studies, benchmarks, and position papers about agents that design, optimize, supervise, train, or improve other agents. Submissions must present original, unpublished work that has not appeared at NeurIPS or other archival machine-learning venues.\n\nKey Dates\nSubmission Deadline: August 29, 2026, AoE\nNotification: on or before September 29, 2026, AoE\nWorkshop Date: December 11-12, 2026 (Sydney)\nAll deadlines follow the Anywhere on Earth (AoE) timezone.\n\nSubmission Site\nSubmissions are managed via OpenReview and remain private during review. All authors should maintain up-to-date OpenReview profiles for conflict-of-interest management and paper matching. Submit your paper at the OpenReview submission portal.\n\nScope\nWe welcome contributions across the topics above. Accepted papers are presented as posters, with a subset selected for oral or spotlight talks, and we give a Best Paper award and a Best Social Impact Paper award for the work that most thoughtfully addresses the societal implications of meta-agents. The workshop is in person at NeurIPS 2026 in Sydney.\n\nSubmission Guidelines\nFormatting: Submissions must be in English and use the NeurIPS 2026 workshop LaTeX template. Papers are submitted as a single PDF:\n- Full Papers: at most 9 pages (main text)\n- Short Papers: at most 4 pages (main text)\n- Demo Track: live demonstrations of meta-agent systems and tools, presented alongside the poster sessions.\n- Position Papers: on the governance, oversight, and societal impact of meta-agents.\nReferences and appendices do not count toward the page limit, but the main text must be self-contained.\n\nResponsible-Use Statement: Every submission also includes a short responsible-use statement covering the potential societal impacts of the proposed work and suggested mitigations. It is reviewed with the paper, and a missing one is grounds for desk rejection.\n\nAnonymity: The workshop uses double-blind review. Submissions must be anonymized, with author names, affiliations, and acknowledgments removed and prior work cited in the third person.\n\nNon-Archival Policy: The workshop is non-archival. Papers under review elsewhere are welcome, and accepted papers may be published at other venues afterward. Work already published at NeurIPS or other archival ML venues should not be submitted.\n\nContact: Email meta-agents-workshop@googlegroups.com." }, { "key": "JUDGe", "title": "First Workshop on Reliable Evaluation for Language Models (JUDGe)", "subtitle": "JUDGe 2026", "summary": "A full-day workshop on building reliable, valid, and robust LLM-based evaluators, treating evaluator reliability and validity as a systems problem that spans upstream and downstream pipeline components.", "cfp_full": "JUDGe @ NeurIPS 2026 - Can We Trust the Judge?\nNeurIPS 2026 · Atlanta · Dec 12-13\n\n\"Rigorous evaluation is the backbone of trustworthy AI - let's scrutinize the scrutinizers.\"\n\nA full-day workshop on building reliable, valid, and robust LLM-based evaluators. We bring together NLP researchers, ML systems builders, safety scientists, and industry practitioners around a single foundational question: how do we know whether an LLM evaluator is actually measuring what we intend it to measure?\n\nAbout the Workshop\nWhy JUDGe?\nEvaluation validity is not a property of a judge in isolation - it is a property of a judge in a system. A well-calibrated evaluator can fail systematically when deployed in a pipeline where its outputs gate safety decisions, feed back into training, or depend on context it was never designed to handle. The field has treated evaluation as a measurement problem - how accurate is the judge? - when the harder question is infrastructural: how does a judge's error profile interact with what is upstream and downstream of it, and what happens downstream when it fails?\n\nJUDGe is the first NeurIPS workshop to take evaluator reliability and validity seriously as a systems problem. Judges embedded in RLHF, DPO, and Constitutional AI pipelines don't just mismeasure - their failure modes compound into model weights and downstream decisions. This workshop is where NLP evaluation researchers, alignment scientists, and production ML practitioners come together around that shared problem.\n\nThemes:\n- Core Evaluation Validity: Construct validity, calibration, human-model alignment, criteria pre-registration, and inter-judge consistency.\n- Surface vs. Semantic Sensitivity: Robustness of LLM judges to meaning-preserving paraphrase, length and formatting bias, and semantic content scoring.\n- Bias & Sycophancy: Positional and ordering bias in pairwise evaluation, self-preference, and feedback loop risks in judge-guided alignment.\n- Safety-Relevant Evaluation: Semantic drift under iterative transformation, adversarial robustness of safety evaluators, and meaning preservation.\n- Benchmark Construction: Adversarial benchmark design, criteria drift in datasets, domain-specific frameworks, and negative-result datasets.\n- Judge Design & Architecture: Prompt engineering, fine-tuning vs. prompting, multi-judge ensembles, tool-augmented judges, and capability gaps.\n- Practitioner Perspectives: Production evaluation pipelines, practitioner-researcher gaps, deployment failure case studies, and cost-quality trade-offs.\n- Agentic & Emerging Topics: Reasoning chain validity, multi-turn agentic evaluation, cross-lingual reliability, and ethical dimensions of automated evaluation.\n\nFailure Taxonomy\nLLM judge failures don't arrive in isolation - in production pipelines they cascade. Sycophancy biases preference data, which shifts model style, which drifts the judge's implicit criteria across training iterations. Surface sensitivity enables adversarial safety bypasses. Correlated errors across judge families mean ensembles suppress disagreement precisely on the cases most likely to be collectively wrong. The taxonomy organizes these into seven empirically grounded failure facets:\n01 Surface vs. semantic sensitivity - Formatting and length drive scores over meaning; paraphrases receive inconsistent judgments. Open question: Can judges be calibrated to score meaning-equivalent responses identically?\n02 Criteria drift - Criteria shift after seeing real outputs; \"evaluate helpfulness\" is operationalized inconsistently. Open question: Can criteria be pre-registered and verified for post-hoc consistency?\n03 Positional & ordering bias - Primacy/recency effects skew pairwise rankings. Open question: Does position-swap averaging fully debias long-context evaluation?\n04 Sycophancy & self-preference - Judges favour stylistically familiar outputs regardless of quality. Open question: How do we detect and break the sycophancy-training feedback loop?\n05 Reasoning chain validity - Judge capability bounds evaluation capability; gap worsens as models outpace judges. Open question: What is the minimum judge-model capability gap for valid evaluation?\n06 Safety-relevant meaning preservation - Judges are fooled by adversarial paraphrase preserving harmful content. Open question: What protocols reliably detect safety-relevant semantic drift?\n07 Inter-judge consistency - Cross-judge agreement is low on semantically complex cases. Open question: What inter-judge agreement threshold is acceptable in high-stakes settings?\n\nCommunity Deliverable: Judge Deployment Disclosure Template\nOne concrete output of JUDGe is a structured disclosure template for judge deployment decisions - analogous to a model card, but for the evaluation pipeline. No such standard currently exists. The organizing team is drafting a seed version from collective production experience at Meta, Amazon, and Google, releasing it on GitHub before the workshop, and refining it collaboratively with attendees during the poster session and panel. The template covers four dimensions: Training Data Provenance; Deployment Context; Known Failure Modes; Human Validation.\n\nCall for Papers\nJUDGe welcomes original research on all dimensions of LLM evaluator reliability and validity. Works in progress, negative results, practitioner case studies, and cross-disciplinary contributions are particularly encouraged - the workshop is designed for work that wouldn't fit neatly into a standard NLP or ML venue track.\n\nSubmission Tracks\n- Full Papers: 6 pages + references · Oral presentation\n- Short Papers: 4 pages + references · Poster presentation\n- Junior Spotlight: 2 pages + references · Oral · Students & early-career only\n\nTopics of Interest\n- Construct validity in LLM-based evaluators\n- Calibration methods for LLM judges\n- Human-model alignment in evaluation\n- Robustness to meaning-preserving paraphrase\n- Positional and ordering bias in pairwise evaluation\n- Sycophancy and self-preference detection\n- Semantic drift detection under iterative transformation\n- Adversarial robustness of safety evaluators\n- Adversarial benchmark design for stress-testing\n- Multi-judge ensembles and disagreement resolution\n- Tool-augmented and environment-grounded judges\n- Production evaluation pipeline case studies\n- Reasoning chain validity in complex task evaluation\n- Agentic and multi-turn LLM evaluation\n- Cross-lingual and multilingual evaluation reliability\n- Societal and ethical dimensions of automated evaluation\n\nAll submissions via OpenReview, double-blind, >=3 reviews per paper. All accepted work is non-archival and posted on the workshop website (authors may opt out). Previously published work at a major ML venue is not eligible.\n\nImportant Dates (all deadlines are 11:59 PM AoE; all notifications precede the mandatory NeurIPS deadline of September 29, 2026)\nCFP Opens: August 1, 2026\nSubmission Deadline: August 29, 2026\nNotifications: September 29, 2026\nCamera-Ready: October 15, 2026\nWorkshop Day: Dec 12-13, 2026 · NeurIPS 2026 · Atlanta, Georgia", "cfp_status": "published", "topics": [ "Construct validity in LLM-based evaluators", "Calibration methods for LLM judges", "Human-model alignment in evaluation", "Robustness to meaning-preserving paraphrase", "Positional and ordering bias in pairwise evaluation", "Sycophancy and self-preference detection", "Semantic drift detection under iterative transformation", "Adversarial robustness of safety evaluators", "Adversarial benchmark design for stress-testing", "Multi-judge ensembles and disagreement resolution", "Tool-augmented and environment-grounded judges", "Production evaluation pipeline case studies", "Reasoning chain validity in complex task evaluation", "Agentic and multi-turn LLM evaluation", "Cross-lingual and multilingual evaluation reliability", "Societal and ethical dimensions of automated evaluation" ], "important_dates": [ { "label": "CFP Opens", "date": "2026-08-01" }, { "label": "Submission Deadline", "date": "2026-08-29" }, { "label": "Notifications", "date": "2026-09-29" }, { "label": "Camera-Ready", "date": "2026-10-15" }, { "label": "Workshop Day", "date": "2026-12-12 to 2026-12-13" } ], "organizers": [ "Dr. Shanu Sushmita (Northeastern University (Seattle) & University of Washington, Lead Organizer)", "Dr. Jayash Koshal (Meta & Northeastern University)", "Meghana Makhija (Amazon)", "Dr. Hui Wan (Google DeepMind)", "Dr. Amjad Abu-Jbara (Amazon Ads)" ], "speakers": [ "Eugene Yan (Anthropic, Keynote)", "Jennifer Wortman Vaughan (Microsoft Research NYC, Panelist)", "Wei Xu (Georgia Tech, Panelist)", "Rui Song (Amazon Core AI, Panelist)" ], "host_url": "https://judge2026.github.io", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/JUDGe", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/JUDGe", "location": "Atlanta, USA", "city": "Atlanta", "workshop_date": "2026-12-12", "contact": "judge-neurips-2026@googlegroups.com", "tracks": [ { "key": "JUDGe", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/JUDGe", "submission_dates_raw": "" } ], "group": "trust", "group_label": "Trustworthy & Safe AI", "corpus": "First Workshop on Reliable Evaluation for Language Models (JUDGe) JUDGe 2026 A full-day workshop on building reliable, valid, and robust LLM-based evaluators, treating evaluator reliability and validity as a systems problem that spans upstream and downstream pipeline components. Construct validity in LLM-based evaluators Calibration methods for LLM judges Human-model alignment in evaluation Robustness to meaning-preserving paraphrase Positional and ordering bias in pairwise evaluation Sycophancy and self-preference detection Semantic drift detection under iterative transformation Adversarial robustness of safety evaluators Adversarial benchmark design for stress-testing Multi-judge ensembles and disagreement resolution Tool-augmented and environment-grounded judges Production evaluation pipeline case studies Reasoning chain validity in complex task evaluation Agentic and multi-turn LLM evaluation Cross-lingual and multilingual evaluation reliability Societal and ethical dimensions of automated evaluation JUDGe @ NeurIPS 2026 - Can We Trust the Judge?\nNeurIPS 2026 · Atlanta · Dec 12-13\n\n\"Rigorous evaluation is the backbone of trustworthy AI - let's scrutinize the scrutinizers.\"\n\nA full-day workshop on building reliable, valid, and robust LLM-based evaluators. We bring together NLP researchers, ML systems builders, safety scientists, and industry practitioners around a single foundational question: how do we know whether an LLM evaluator is actually measuring what we intend it to measure?\n\nAbout the Workshop\nWhy JUDGe?\nEvaluation validity is not a property of a judge in isolation - it is a property of a judge in a system. A well-calibrated evaluator can fail systematically when deployed in a pipeline where its outputs gate safety decisions, feed back into training, or depend on context it was never designed to handle. The field has treated evaluation as a measurement problem - how accurate is the judge? - when the harder question is infrastructural: how does a judge's error profile interact with what is upstream and downstream of it, and what happens downstream when it fails?\n\nJUDGe is the first NeurIPS workshop to take evaluator reliability and validity seriously as a systems problem. Judges embedded in RLHF, DPO, and Constitutional AI pipelines don't just mismeasure - their failure modes compound into model weights and downstream decisions. This workshop is where NLP evaluation researchers, alignment scientists, and production ML practitioners come together around that shared problem.\n\nThemes:\n- Core Evaluation Validity: Construct validity, calibration, human-model alignment, criteria pre-registration, and inter-judge consistency.\n- Surface vs. Semantic Sensitivity: Robustness of LLM judges to meaning-preserving paraphrase, length and formatting bias, and semantic content scoring.\n- Bias & Sycophancy: Positional and ordering bias in pairwise evaluation, self-preference, and feedback loop risks in judge-guided alignment.\n- Safety-Relevant Evaluation: Semantic drift under iterative transformation, adversarial robustness of safety evaluators, and meaning preservation.\n- Benchmark Construction: Adversarial benchmark design, criteria drift in datasets, domain-specific frameworks, and negative-result datasets.\n- Judge Design & Architecture: Prompt engineering, fine-tuning vs. prompting, multi-judge ensembles, tool-augmented judges, and capability gaps.\n- Practitioner Perspectives: Production evaluation pipelines, practitioner-researcher gaps, deployment failure case studies, and cost-quality trade-offs.\n- Agentic & Emerging Topics: Reasoning chain validity, multi-turn agentic evaluation, cross-lingual reliability, and ethical dimensions of automated evaluation.\n\nFailure Taxonomy\nLLM judge failures don't arrive in isolation - in production pipelines they cascade. Sycophancy biases preference data, which shifts model style, which drifts the judge's implicit criteria across training iterations. Surface sensitivity enables adversarial safety bypasses. Correlated errors across judge families mean ensembles suppress disagreement precisely on the cases most likely to be collectively wrong. The taxonomy organizes these into seven empirically grounded failure facets:\n01 Surface vs. semantic sensitivity - Formatting and length drive scores over meaning; paraphrases receive inconsistent judgments. Open question: Can judges be calibrated to score meaning-equivalent responses identically?\n02 Criteria drift - Criteria shift after seeing real outputs; \"evaluate helpfulness\" is operationalized inconsistently. Open question: Can criteria be pre-registered and verified for post-hoc consistency?\n03 Positional & ordering bias - Primacy/recency effects skew pairwise rankings. Open question: Does position-swap averaging fully debias long-context evaluation?\n04 Sycophancy & self-preference - Judges favour stylistically familiar outputs regardless of quality. Open question: How do we detect and break the sycophancy-training feedback loop?\n05 Reasoning chain validity - Judge capability bounds evaluation capability; gap worsens as models outpace judges. Open question: What is the minimum judge-model capability gap for valid evaluation?\n06 Safety-relevant meaning preservation - Judges are fooled by adversarial paraphrase preserving harmful content. Open question: What protocols reliably detect safety-relevant semantic drift?\n07 Inter-judge consistency - Cross-judge agreement is low on semantically complex cases. Open question: What inter-judge agreement threshold is acceptable in high-stakes settings?\n\nCommunity Deliverable: Judge Deployment Disclosure Template\nOne concrete output of JUDGe is a structured disclosure template for judge deployment decisions - analogous to a model card, but for the evaluation pipeline. No such standard currently exists. The organizing team is drafting a seed version from collective production experience at Meta, Amazon, and Google, releasing it on GitHub before the workshop, and refining it collaboratively with attendees during the poster session and panel. The template covers four dimensions: Training Data Provenance; Deployment Context; Known Failure Modes; Human Validation.\n\nCall for Papers\nJUDGe welcomes original research on all dimensions of LLM evaluator reliability and validity. Works in progress, negative results, practitioner case studies, and cross-disciplinary contributions are particularly encouraged - the workshop is designed for work that wouldn't fit neatly into a standard NLP or ML venue track.\n\nSubmission Tracks\n- Full Papers: 6 pages + references · Oral presentation\n- Short Papers: 4 pages + references · Poster presentation\n- Junior Spotlight: 2 pages + references · Oral · Students & early-career only\n\nTopics of Interest\n- Construct validity in LLM-based evaluators\n- Calibration methods for LLM judges\n- Human-model alignment in evaluation\n- Robustness to meaning-preserving paraphrase\n- Positional and ordering bias in pairwise evaluation\n- Sycophancy and self-preference detection\n- Semantic drift detection under iterative transformation\n- Adversarial robustness of safety evaluators\n- Adversarial benchmark design for stress-testing\n- Multi-judge ensembles and disagreement resolution\n- Tool-augmented and environment-grounded judges\n- Production evaluation pipeline case studies\n- Reasoning chain validity in complex task evaluation\n- Agentic and multi-turn LLM evaluation\n- Cross-lingual and multilingual evaluation reliability\n- Societal and ethical dimensions of automated evaluation\n\nAll submissions via OpenReview, double-blind, >=3 reviews per paper. All accepted work is non-archival and posted on the workshop website (authors may opt out). Previously published work at a major ML venue is not eligible.\n\nImportant Dates (all deadlines are 11:59 PM AoE; all notifications precede the mandatory NeurIPS deadline of September 29, 2026)\nCFP Opens: August 1, 2026\nSubmission Deadline: August 29, 2026\nNotifications: September 29, 2026\nCamera-Ready: October 15, 2026\nWorkshop Day: Dec 12-13, 2026 · NeurIPS 2026 · Atlanta, Georgia" }, { "key": "FAST", "title": "Foundations of Agentic Systems Theory", "subtitle": "FAST 2026", "summary": "The Foundations of Agentic Systems Theory (FAST) workshop investigates how existing theory (notably from beyond the traditional AI community, including complex systems, developmental biology, organizational sociology, and cognitive science) and new insights unique to LLM-based agents can build understanding of the system-level behavior and risks of agentic AI systems.", "cfp_full": "Foundations of Agentic Systems Theory\nFAST @ NeurIPS 2026, Paris\n\nAs with any complex system, the most interesting and consequential behaviors often arise not from the parts in isolation, but from the patterns of interaction between them. The current development of agentic AI has largely ignored these considerations, instead focusing on designing more (individually) capable agents. Failing to consider these effects as AI agents become more widespread will lead to a significant underestimation in both their capabilities and risks.\nThere is an extensive body of knowledge underlying these interaction effects across various fields, but it's not currently clear how applicable existing theoretical tools are to agentic AI systems. Tools from control theory, game/economic theory, and operations research typically impose strong structural assumptions on both agents and the overall system (such as the form of objective functions, state evolution/dynamics, or degree of rationality) in efforts to obtain concrete results. On the other hand, methods from the social sciences use observations of human behavior, cultural contexts, and social norms to make more measured claims about probable patterns within the complexity and variability of human experience. Agentic AI systems don't cleanly map to either of these settings. The underlying LLM in an AI agent does not possess the same rational behavior as idealized control/game/economic agents, nor does it exhibit the culturally/emotionally/evolutionarily shaped behaviors that characterize human agents.\nThe Foundations of Agentic Systems Theory (FAST) workshop provides a venue for this investigation. Drawing from a variety of fields (notably beyond computer science, including complex systems, developmental biology, organizational sociology, and cognitive science), FAST explores which mechanisms of emergent behavior from other systems carry over to systems of LLM-based agents, the properties of the underlying agents (and their LLMs) that facilitate or impede these behaviors, and the extent to which system-wide outcomes can be controlled or induced. We strongly seek interdisciplinary participation (via both contributions and invited talks), with the ultimate goal of fundamentally contributing to a better understanding of the underlying processes that govern the system-level behavior (and risks) of agentic AI.\n\nScope and Topics\nLarge language models have recently become sophisticated enough to be reliably integrated into more complex pipelines, leading to more automated (i.e., agentic) use cases. However, the community has focused disproportionately on building these systems rather than understanding why they may (or may not) work. The goal of the FAST workshop is to investigate how both existing theory (notably that outside of the traditional AI community) and new insights (unique to LLM-based agents) can help to build this understanding.\nAs such, we invite submissions on the following topics:\n- Mechanisms of emergent capabilities and behaviors in agentic systems\n- Evaluation, detection, and bounding of emergent capabilities or failure modes\n- Harness engineering and (neuro-symbolic) scaffolding\n- Theory of mind and recursive social cognition\n- Formation of norms, conventions, and collective bias in populations of agents\n- Compositional safety and governance\n- Definitions and philosophy of agency and emergence in engineered systems\n- Observability/monitorability and steerability/controllability of populations of agents\n\nSubmission Information\nSubmissions can be either full or short papers:\nFull papers: Up to 7 pages (excluding references and appendices); should present mature or completed research.\nShort papers: Up to 4 pages (excluding references and appendices); intended for describing ongoing work, early-stage ideas, or the release of benchmarks and datasets (authors are encouraged to use the short paper format for benchmarks and datasets).\nAll submissions must be made through our OpenReview page. Please use the NeurIPS 2026 template when preparing your submission.\nSubmissions must be anonymized for double-blind review. Reviewing will follow the standards of NeurIPS, with evaluation based on novelty, technical depth, clarity, reproducibility, and potential impact. Accepted papers will be presented as either posters or contributed talks. At least one author of each accepted paper must register and attend the workshop. If you have any questions, please contact us at fast.workshop.team@gmail.com.\n\nImportant Dates\nSubmission window opens (OpenReview): TBA\nPaper submission deadline: August 29, 2026 (AoE)\nAcceptance notification: September 26, 2026 (AoE)\nWorkshop (exact day to be announced): December 12 or 13, 2026\n\nPanel Discussion\nThe day will feature an hour-long moderated panel, \"Mind to machines: Emergence, methods, and risks\" (moderated by Joshua Krook), consisting of selected experts in the field. The panel members and the specific topics for discussion will be announced closer to the workshop.", "cfp_status": "published", "topics": [ "Mechanisms of emergent capabilities and behaviors in agentic systems", "Evaluation, detection, and bounding of emergent capabilities or failure modes", "Harness engineering and (neuro-symbolic) scaffolding", "Theory of mind and recursive social cognition", "Formation of norms, conventions, and collective bias in populations of agents", "Compositional safety and governance", "Definitions and philosophy of agency and emergence in engineered systems", "Observability/monitorability and steerability/controllability of populations of agents" ], "important_dates": [ { "label": "Submission window opens (OpenReview)", "date": "TBA" }, { "label": "Paper submission deadline", "date": "2026-08-29" }, { "label": "Acceptance notification", "date": "2026-09-26" }, { "label": "Workshop", "date": "December 12 or 13, 2026" } ], "organizers": [ "Erik Miehling (IBM Research - Ireland)", "Madeline Reinecke (University of Oxford)", "Hamza Mostafa (University of Waterloo)", "Jordan McAfoose (IBM Research - Zurich)", "Irina Rish (Université de Montréal / Mila)", "Djallel Bouneffouf (IBM Research - Yorktown Heights)" ], "speakers": [ "Michael Levin (Tufts University)", "Danielle Perszyk (Amazon AGI SF Lab)", "Winnie Street (Google Research)", "Atoosa Kasirzadeh (Carnegie Mellon University / Google DeepMind)" ], "host_url": "https://fast-workshop.github.io", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/FAST", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/FAST", "location": "Paris, France", "city": "Paris", "workshop_date": "2026-12-12", "contact": "erik.miehling@ibm.com", "tracks": [ { "key": "FAST", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/FAST", "submission_dates_raw": "" } ], "group": "agents", "group_label": "Agents & Agentic Systems", "corpus": "Foundations of Agentic Systems Theory FAST 2026 The Foundations of Agentic Systems Theory (FAST) workshop investigates how existing theory (notably from beyond the traditional AI community, including complex systems, developmental biology, organizational sociology, and cognitive science) and new insights unique to LLM-based agents can build understanding of the system-level behavior and risks of agentic AI systems. Mechanisms of emergent capabilities and behaviors in agentic systems Evaluation, detection, and bounding of emergent capabilities or failure modes Harness engineering and (neuro-symbolic) scaffolding Theory of mind and recursive social cognition Formation of norms, conventions, and collective bias in populations of agents Compositional safety and governance Definitions and philosophy of agency and emergence in engineered systems Observability/monitorability and steerability/controllability of populations of agents Foundations of Agentic Systems Theory\nFAST @ NeurIPS 2026, Paris\n\nAs with any complex system, the most interesting and consequential behaviors often arise not from the parts in isolation, but from the patterns of interaction between them. The current development of agentic AI has largely ignored these considerations, instead focusing on designing more (individually) capable agents. Failing to consider these effects as AI agents become more widespread will lead to a significant underestimation in both their capabilities and risks.\nThere is an extensive body of knowledge underlying these interaction effects across various fields, but it's not currently clear how applicable existing theoretical tools are to agentic AI systems. Tools from control theory, game/economic theory, and operations research typically impose strong structural assumptions on both agents and the overall system (such as the form of objective functions, state evolution/dynamics, or degree of rationality) in efforts to obtain concrete results. On the other hand, methods from the social sciences use observations of human behavior, cultural contexts, and social norms to make more measured claims about probable patterns within the complexity and variability of human experience. Agentic AI systems don't cleanly map to either of these settings. The underlying LLM in an AI agent does not possess the same rational behavior as idealized control/game/economic agents, nor does it exhibit the culturally/emotionally/evolutionarily shaped behaviors that characterize human agents.\nThe Foundations of Agentic Systems Theory (FAST) workshop provides a venue for this investigation. Drawing from a variety of fields (notably beyond computer science, including complex systems, developmental biology, organizational sociology, and cognitive science), FAST explores which mechanisms of emergent behavior from other systems carry over to systems of LLM-based agents, the properties of the underlying agents (and their LLMs) that facilitate or impede these behaviors, and the extent to which system-wide outcomes can be controlled or induced. We strongly seek interdisciplinary participation (via both contributions and invited talks), with the ultimate goal of fundamentally contributing to a better understanding of the underlying processes that govern the system-level behavior (and risks) of agentic AI.\n\nScope and Topics\nLarge language models have recently become sophisticated enough to be reliably integrated into more complex pipelines, leading to more automated (i.e., agentic) use cases. However, the community has focused disproportionately on building these systems rather than understanding why they may (or may not) work. The goal of the FAST workshop is to investigate how both existing theory (notably that outside of the traditional AI community) and new insights (unique to LLM-based agents) can help to build this understanding.\nAs such, we invite submissions on the following topics:\n- Mechanisms of emergent capabilities and behaviors in agentic systems\n- Evaluation, detection, and bounding of emergent capabilities or failure modes\n- Harness engineering and (neuro-symbolic) scaffolding\n- Theory of mind and recursive social cognition\n- Formation of norms, conventions, and collective bias in populations of agents\n- Compositional safety and governance\n- Definitions and philosophy of agency and emergence in engineered systems\n- Observability/monitorability and steerability/controllability of populations of agents\n\nSubmission Information\nSubmissions can be either full or short papers:\nFull papers: Up to 7 pages (excluding references and appendices); should present mature or completed research.\nShort papers: Up to 4 pages (excluding references and appendices); intended for describing ongoing work, early-stage ideas, or the release of benchmarks and datasets (authors are encouraged to use the short paper format for benchmarks and datasets).\nAll submissions must be made through our OpenReview page. Please use the NeurIPS 2026 template when preparing your submission.\nSubmissions must be anonymized for double-blind review. Reviewing will follow the standards of NeurIPS, with evaluation based on novelty, technical depth, clarity, reproducibility, and potential impact. Accepted papers will be presented as either posters or contributed talks. At least one author of each accepted paper must register and attend the workshop. If you have any questions, please contact us at fast.workshop.team@gmail.com.\n\nImportant Dates\nSubmission window opens (OpenReview): TBA\nPaper submission deadline: August 29, 2026 (AoE)\nAcceptance notification: September 26, 2026 (AoE)\nWorkshop (exact day to be announced): December 12 or 13, 2026\n\nPanel Discussion\nThe day will feature an hour-long moderated panel, \"Mind to machines: Emergence, methods, and risks\" (moderated by Joshua Krook), consisting of selected experts in the field. The panel members and the specific topics for discussion will be announced closer to the workshop." }, { "key": "FLMSec", "title": "Foundations of Language Model Security: Theory, Practice, and Fundamental Limits", "subtitle": "FLMSec 2026", "summary": "A NeurIPS 2026 workshop advancing research on secure-by-design LLM systems, shifting away from the cat-and-mouse game of attacks and defenses toward a principled understanding of why security vulnerabilities arise and how to address them from the ground up, organized around formalizing LLM security, security in practice, and fundamental limits.", "cfp_full": "About the Workshop\nSecure LLM systems by design before insecure patterns become the default.\n\nThis workshop aims to advance research on secure-by-design LLM systems by shifting away from the current cat-and-mouse game of attacks and defenses toward a principled understanding of why security vulnerabilities arise and how to address them from the ground up. LLMs have been shown to be vulnerable to a range of attacks such as prompt injections and data poisoning, and yet continue to be deployed in complex systems without a clear understanding of why these vulnerabilities arise or how they interconnect with classical security vulnerabilities.\n\nAt the model level, the absence of a hard separation between instructions and data may expose fundamental attack surfaces; at the system level, confused-deputy patterns and missing trust boundaries introduce further structural weaknesses. Understanding whether these vulnerabilities are inherent to current language modeling architectures or artifacts of specific design choices is essential for moving from brittle empirical defenses toward principled security.\n\nRecent research advocates treating LLM security as a system design problem, with a few approaches achieving provable security in specific settings. However, the field still lacks shared formal definitions of what LLM security means, comparable to differential privacy for privacy guarantees, and securing systems by design remains a use-case-specific engineering effort rather than an application of generic principles.\n\nMoreover, existing solutions that offer security guarantees tend to degrade the utility of the system, and it is unclear whether this trade-off is an artifact of current approaches or a more fundamental limit of any LLM system that achieves meaningful security. This workshop aims to consolidate existing knowledge and lay the foundations for future LLM security research by answering three questions:\n\nQ1 — Formalizing LLM Security: How should LLM security be formalized? Is there an agreed-upon framework comparable to the notion of differential privacy in privacy research? What role should theory play in creating secure LLM systems?\n\nQ2 — Security in Practice: Can we design evaluation methodologies that are reproducible and generalizable rather than fragile and hackable? What concrete steps can help avoid unproductive cycles of attacks and defenses?\n\nQ3 — Fundamental Limits of Security: Existing approaches to securing LLM-based systems trade off security for utility. Is this trade-off an artifact of current defense designs, or a more fundamental property that any secure system must exhibit? In which settings has provable security already been achieved, and are there impossibility results establishing conditions under which security cannot be attained?\n\nWorkshop Format\nThe workshop will consist of four thematic blocks. Each block will consist of a 45-minute expert keynote followed by a 30-minute guided discussion, in which participants split into groups of 10-15 with an assigned discussion chair. The goal of the discussion is to identify open questions that emerge from the keynote. At the end of the discussion, each chair will deliver a 2-minute summary to all attendees. The program will also include a joint poster session, two spotlight contributed talks, and interactive demos to encourage discussion beyond the themes of each block.\n\nCall for Papers\nWe invite non-archival submissions of up to 8 pages (excluding references) in the NeurIPS workshop template. We will use OpenReview to manage submissions, as it allows us to streamline notifications and automatically enforce NeurIPS conflict-of-interest policies. Submissions will be managed through OpenReview. All accepted papers will be presented as posters, with one to two selected for short talks and receiving a best paper award.\n\nTopics of Interest\n- Formal frameworks and definitions for LLM security [Q1]\n- Secure-by-design LLM system architectures [Q1, Q2]\n- Provable security results and impossibility results for LLM-based systems [Q3]\n- Design of reproducible, generalizable evaluation methodologies [Q2]\n- The security–utility trade-off in current and future defenses [Q3]\n- Model- and system-level attacks and defenses (prompt injection, data poisoning, jailbreaks) situated within a principled security framework [Q1, Q2]\n- Multi-agent security and worst-case security guarantees [Q2, Q3]\n- Memorization and privacy in language models [Q1, Q3]\n- Compositional security: do component-level guarantees hold when models or agents are composed into larger systems? [Q3]\n\nSubmission Policy\nIn accordance with NeurIPS policy, individuals with a personal conflict of interest with any organizer are not permitted to submit to the workshop. Previously published work is not eligible. We recommend all authors register on OpenReview at least two weeks before the submission deadline.\n\nImportant Dates\nAll deadlines are 23:59 AoE (Anywhere on Earth).\n- Submissions open: July 23, 2026\n- Submission deadline: August 22, 2026\n- Reviewer bidding: August 22 – 25, 2026\n- Reviewing period: August 25 – September 25, 2026\n- Decisions: September 25 – 29, 2026", "cfp_status": "published", "topics": [ "Formal frameworks and definitions for LLM security", "Secure-by-design LLM system architectures", "Provable security results and impossibility results for LLM-based systems", "Design of reproducible, generalizable evaluation methodologies", "The security–utility trade-off in current and future defenses", "Model- and system-level attacks and defenses (prompt injection, data poisoning, jailbreaks) within a principled security framework", "Multi-agent security and worst-case security guarantees", "Memorization and privacy in language models", "Compositional security of composed models and agents" ], "important_dates": [ { "label": "Submissions Open", "date": "2026-07-23" }, { "label": "Submission Deadline", "date": "2026-08-22" }, { "label": "Reviewer Bidding", "date": "2026-08-22 to 2026-08-25" }, { "label": "Reviewing Period", "date": "2026-08-25 to 2026-09-25" }, { "label": "Decisions", "date": "2026-09-25 to 2026-09-29" } ], "organizers": [ "Egor Zverev (Institute of Science and Technology Austria)", "Maura Pintor (University of Cagliari)", "Santiago Zanella-Béguelin (Microsoft)", "Ana-Maria Cretu (CISPA Helmholtz Center for Information Security)", "Nicole Nichols (Palo Alto Networks)", "Pavel Laskov (University of Liechtenstein)" ], "speakers": [ "Niloofar Mireshghallah (humans& and CMU)", "Christian Schroeder de Witt (University of Oxford)", "Reza Shokri (Google and National University of Singapore)", "Somesh Jha (University of Wisconsin-Madison)", "Julia Bazinska (Lakera AI) — Demo Presenter", "Matthew Maisel (Sondera) — Demo Presenter" ], "host_url": "https://flmsec.github.io/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/FLMSec", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/FLMSec", "location": "Paris, France", "city": "Paris", "workshop_date": "2026-12-12", "contact": "maura.pintor@unica.it", "tracks": [ { "key": "FLMSec", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/FLMSec", "submission_dates_raw": "" } ], "group": "trust", "group_label": "Trustworthy & Safe AI", "corpus": "Foundations of Language Model Security: Theory, Practice, and Fundamental Limits FLMSec 2026 A NeurIPS 2026 workshop advancing research on secure-by-design LLM systems, shifting away from the cat-and-mouse game of attacks and defenses toward a principled understanding of why security vulnerabilities arise and how to address them from the ground up, organized around formalizing LLM security, security in practice, and fundamental limits. Formal frameworks and definitions for LLM security Secure-by-design LLM system architectures Provable security results and impossibility results for LLM-based systems Design of reproducible, generalizable evaluation methodologies The security–utility trade-off in current and future defenses Model- and system-level attacks and defenses (prompt injection, data poisoning, jailbreaks) within a principled security framework Multi-agent security and worst-case security guarantees Memorization and privacy in language models Compositional security of composed models and agents About the Workshop\nSecure LLM systems by design before insecure patterns become the default.\n\nThis workshop aims to advance research on secure-by-design LLM systems by shifting away from the current cat-and-mouse game of attacks and defenses toward a principled understanding of why security vulnerabilities arise and how to address them from the ground up. LLMs have been shown to be vulnerable to a range of attacks such as prompt injections and data poisoning, and yet continue to be deployed in complex systems without a clear understanding of why these vulnerabilities arise or how they interconnect with classical security vulnerabilities.\n\nAt the model level, the absence of a hard separation between instructions and data may expose fundamental attack surfaces; at the system level, confused-deputy patterns and missing trust boundaries introduce further structural weaknesses. Understanding whether these vulnerabilities are inherent to current language modeling architectures or artifacts of specific design choices is essential for moving from brittle empirical defenses toward principled security.\n\nRecent research advocates treating LLM security as a system design problem, with a few approaches achieving provable security in specific settings. However, the field still lacks shared formal definitions of what LLM security means, comparable to differential privacy for privacy guarantees, and securing systems by design remains a use-case-specific engineering effort rather than an application of generic principles.\n\nMoreover, existing solutions that offer security guarantees tend to degrade the utility of the system, and it is unclear whether this trade-off is an artifact of current approaches or a more fundamental limit of any LLM system that achieves meaningful security. This workshop aims to consolidate existing knowledge and lay the foundations for future LLM security research by answering three questions:\n\nQ1 — Formalizing LLM Security: How should LLM security be formalized? Is there an agreed-upon framework comparable to the notion of differential privacy in privacy research? What role should theory play in creating secure LLM systems?\n\nQ2 — Security in Practice: Can we design evaluation methodologies that are reproducible and generalizable rather than fragile and hackable? What concrete steps can help avoid unproductive cycles of attacks and defenses?\n\nQ3 — Fundamental Limits of Security: Existing approaches to securing LLM-based systems trade off security for utility. Is this trade-off an artifact of current defense designs, or a more fundamental property that any secure system must exhibit? In which settings has provable security already been achieved, and are there impossibility results establishing conditions under which security cannot be attained?\n\nWorkshop Format\nThe workshop will consist of four thematic blocks. Each block will consist of a 45-minute expert keynote followed by a 30-minute guided discussion, in which participants split into groups of 10-15 with an assigned discussion chair. The goal of the discussion is to identify open questions that emerge from the keynote. At the end of the discussion, each chair will deliver a 2-minute summary to all attendees. The program will also include a joint poster session, two spotlight contributed talks, and interactive demos to encourage discussion beyond the themes of each block.\n\nCall for Papers\nWe invite non-archival submissions of up to 8 pages (excluding references) in the NeurIPS workshop template. We will use OpenReview to manage submissions, as it allows us to streamline notifications and automatically enforce NeurIPS conflict-of-interest policies. Submissions will be managed through OpenReview. All accepted papers will be presented as posters, with one to two selected for short talks and receiving a best paper award.\n\nTopics of Interest\n- Formal frameworks and definitions for LLM security [Q1]\n- Secure-by-design LLM system architectures [Q1, Q2]\n- Provable security results and impossibility results for LLM-based systems [Q3]\n- Design of reproducible, generalizable evaluation methodologies [Q2]\n- The security–utility trade-off in current and future defenses [Q3]\n- Model- and system-level attacks and defenses (prompt injection, data poisoning, jailbreaks) situated within a principled security framework [Q1, Q2]\n- Multi-agent security and worst-case security guarantees [Q2, Q3]\n- Memorization and privacy in language models [Q1, Q3]\n- Compositional security: do component-level guarantees hold when models or agents are composed into larger systems? [Q3]\n\nSubmission Policy\nIn accordance with NeurIPS policy, individuals with a personal conflict of interest with any organizer are not permitted to submit to the workshop. Previously published work is not eligible. We recommend all authors register on OpenReview at least two weeks before the submission deadline.\n\nImportant Dates\nAll deadlines are 23:59 AoE (Anywhere on Earth).\n- Submissions open: July 23, 2026\n- Submission deadline: August 22, 2026\n- Reviewer bidding: August 22 – 25, 2026\n- Reviewing period: August 25 – September 25, 2026\n- Decisions: September 25 – 29, 2026" }, { "key": "VLM4RWD", "title": "Grounded and Faithful Vision-Language Models for Real-World Deployment 2026", "subtitle": "VLM4RWD2026", "summary": "The 2nd VLM4RWD Workshop focuses on the principles, methods, and evaluation needed to build grounded, faithful, and reliable multimodal intelligence for real-world deployment, spanning visual grounding, faithful reasoning, hallucination mitigation, embodied agents, robotics, and robust evaluation.", "cfp_full": "Grounded and Faithful Vision-Language Models for Real-World Deployment.\n2nd Workshop on VLM4RWD | NeurIPS 2026 | Dec, 2026 | Sydney, Australia\n\nWorkshop Overview\nVision-language(-action) models are rapidly evolving from systems that describe the world to agents that must perceive, reason, and act within it. This workshop focuses on the principles, methods, and evaluation needed to build grounded, faithful, and reliable multimodal intelligence for real-world deployment.\n\nGrounded Understanding & Faithful Reasoning\nAI systems must reliably connect language, perception, and actions with relevant entities and observations in the environment, ensuring that predictions, reasoning, and decisions remain supported by underlying visual and physical evidence. (Visual grounding, Faithful Reasoning, Evidence alignment, Hallucination mitigation)\n\nReliable Interaction & Decision-Making\nReal-world agents must integrate perception, reasoning, planning, and action in a manner that remains robust under dynamic and uncertain conditions, with decisions grounded in the causal structure of the environment. (Embodied AI, Autonomous Systems, Causal Decision-Making)\n\nEvaluation & Deployment Reliability\nSystems deployed in robotics and autonomous environments require principled evaluation, calibrated uncertainty, predictable failure modes, and rigorous assessment across distribution shifts and counterfactual scenarios. (Benchmarks, Uncertainty, Robust Evaluation)\n\nCall for Papers\n\nOverview\nWe invite high-quality submissions that advance grounded and faithful vision-language and vision-language-action models for real-world deployment.\n\nWe welcome research addressing key challenges in visual grounding, faithful reasoning, hallucination mitigation, robustness, embodied intelligence, robotics, autonomous systems, and world models.\n\nAccepted papers will be presented during the workshop poster sessions, and selected submissions will be invited for contributed spotlight talks.\n\nSubmission Guidelines\nFormatting: Workshop papers may be up to 8 pages, excluding references and appendices, and should follow the NeurIPS 2026 conference format.\nReview process: Submissions will undergo double-blind review and must be fully anonymized.\nSubmission portal: Papers will be submitted through OpenReview.\nSubmission types: We welcome full workshop papers, demo papers, extended abstracts, position papers, datasets, benchmarks, and emerging research ideas.\nPreviously published work: Relevant published work may be submitted for presentation but will not be eligible for workshop awards.\n\nTopics of Interest\n- Grounded perception and evidence localization for multimodal systems\n- Faithful reasoning and evidence-grounded decision-making\n- Hallucination detection and mitigation in vision-language(-action) systems\n- Spatial, compositional, temporal, and relational reasoning in grounded multimodal systems\n- Causal reasoning, causal grounding, and counterfactual reasoning in multimodal systems\n- Training paradigms and architectures for grounded VLMs and VLAs\n- Reliable decision-making for embodied agents under dynamic and uncertain conditions\n- Out-of-distribution detection and failure mode analysis for grounded multimodal pipelines\n- Benchmarks, datasets, and evaluation methodologies for grounding and faithfulness\n- Interpretability and explainability of grounded multimodal reasoning\n- Emerging directions in grounded multimodal intelligence: world models and agentic VLMs\n\nImportant Dates\n- Paper Submission: Aug 30th, 2026\n- Notification: Sep 29th, 2026\n- Camera Ready: Oct, 2026\n- Workshop Date: Dec, 2026\n\nFAQ\nIs the workshop in-person or virtual? The workshop will be held in-person at NeurIPS 2026 in Sydney, Australia.\nWill the workshop proceedings be archival? No, the workshop proceedings will be non-archival. Authors of accepted papers retain the full copyright of their work and are free to submit extended versions to conferences or journals.", "cfp_status": "published", "topics": [ "Grounded perception and evidence localization for multimodal systems", "Faithful reasoning and evidence-grounded decision-making", "Hallucination detection and mitigation in vision-language(-action) systems", "Spatial, compositional, temporal, and relational reasoning in grounded multimodal systems", "Causal reasoning, causal grounding, and counterfactual reasoning in multimodal systems", "Training paradigms and architectures for grounded VLMs and VLAs", "Reliable decision-making for embodied agents under dynamic and uncertain conditions", "Out-of-distribution detection and failure mode analysis for grounded multimodal pipelines", "Benchmarks, datasets, and evaluation methodologies for grounding and faithfulness", "Interpretability and explainability of grounded multimodal reasoning", "Emerging directions in grounded multimodal intelligence: world models and agentic VLMs" ], "important_dates": [ { "label": "Paper Submission", "date": "2026-08-30" }, { "label": "Notification", "date": "2026-09-29" }, { "label": "Camera Ready", "date": "2026-10" }, { "label": "Workshop Date", "date": "2026-12" } ], "organizers": [ "Mozhgan Nasr Azadani (University of Waterloo, Stanford)", "Yimu Wang (RBC, University of Waterloo)", "Milan Ganai (Stanford)", "Jiayuan Mao (Amazon FAR, University of Pennsylvania)", "William Zhi (University of Sydney)", "Elahe Arani (Wayve, Eindhoven University of Technology)", "Krzysztof Czarnecki (University of Waterloo)", "Marco Pavone (Stanford, NVIDIA)" ], "speakers": [ "Azalia Mirhoseini (Stanford, Recursive Intelligence)", "Chang Xu (University of Sydney)", "Alexander Toshev (Apple ML Research)", "Igor Gilitschenski (University of Toronto)", "Manling Li (Northwestern University)", "Kashyap Chitta (KESAI)", "Chuchu Fan (MIT)", "Vijay Badrinarayanan (Wayve)", "Kristen Grauman (UT Austin)" ], "host_url": "https://vlm4rwd.github.io", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/VLM4RWD", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/VLM4RWD", "location": "Sydney, Australia", "city": "Sydney", "workshop_date": "2026-12-11", "contact": "vlm4rwd@googlegroups.com", "tracks": [ { "key": "VLM4RWD", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/VLM4RWD", "submission_dates_raw": "" } ], "group": "fm", "group_label": "Foundation Models & LLMs", "corpus": "Grounded and Faithful Vision-Language Models for Real-World Deployment 2026 VLM4RWD2026 The 2nd VLM4RWD Workshop focuses on the principles, methods, and evaluation needed to build grounded, faithful, and reliable multimodal intelligence for real-world deployment, spanning visual grounding, faithful reasoning, hallucination mitigation, embodied agents, robotics, and robust evaluation. Grounded perception and evidence localization for multimodal systems Faithful reasoning and evidence-grounded decision-making Hallucination detection and mitigation in vision-language(-action) systems Spatial, compositional, temporal, and relational reasoning in grounded multimodal systems Causal reasoning, causal grounding, and counterfactual reasoning in multimodal systems Training paradigms and architectures for grounded VLMs and VLAs Reliable decision-making for embodied agents under dynamic and uncertain conditions Out-of-distribution detection and failure mode analysis for grounded multimodal pipelines Benchmarks, datasets, and evaluation methodologies for grounding and faithfulness Interpretability and explainability of grounded multimodal reasoning Emerging directions in grounded multimodal intelligence: world models and agentic VLMs Grounded and Faithful Vision-Language Models for Real-World Deployment.\n2nd Workshop on VLM4RWD | NeurIPS 2026 | Dec, 2026 | Sydney, Australia\n\nWorkshop Overview\nVision-language(-action) models are rapidly evolving from systems that describe the world to agents that must perceive, reason, and act within it. This workshop focuses on the principles, methods, and evaluation needed to build grounded, faithful, and reliable multimodal intelligence for real-world deployment.\n\nGrounded Understanding & Faithful Reasoning\nAI systems must reliably connect language, perception, and actions with relevant entities and observations in the environment, ensuring that predictions, reasoning, and decisions remain supported by underlying visual and physical evidence. (Visual grounding, Faithful Reasoning, Evidence alignment, Hallucination mitigation)\n\nReliable Interaction & Decision-Making\nReal-world agents must integrate perception, reasoning, planning, and action in a manner that remains robust under dynamic and uncertain conditions, with decisions grounded in the causal structure of the environment. (Embodied AI, Autonomous Systems, Causal Decision-Making)\n\nEvaluation & Deployment Reliability\nSystems deployed in robotics and autonomous environments require principled evaluation, calibrated uncertainty, predictable failure modes, and rigorous assessment across distribution shifts and counterfactual scenarios. (Benchmarks, Uncertainty, Robust Evaluation)\n\nCall for Papers\n\nOverview\nWe invite high-quality submissions that advance grounded and faithful vision-language and vision-language-action models for real-world deployment.\n\nWe welcome research addressing key challenges in visual grounding, faithful reasoning, hallucination mitigation, robustness, embodied intelligence, robotics, autonomous systems, and world models.\n\nAccepted papers will be presented during the workshop poster sessions, and selected submissions will be invited for contributed spotlight talks.\n\nSubmission Guidelines\nFormatting: Workshop papers may be up to 8 pages, excluding references and appendices, and should follow the NeurIPS 2026 conference format.\nReview process: Submissions will undergo double-blind review and must be fully anonymized.\nSubmission portal: Papers will be submitted through OpenReview.\nSubmission types: We welcome full workshop papers, demo papers, extended abstracts, position papers, datasets, benchmarks, and emerging research ideas.\nPreviously published work: Relevant published work may be submitted for presentation but will not be eligible for workshop awards.\n\nTopics of Interest\n- Grounded perception and evidence localization for multimodal systems\n- Faithful reasoning and evidence-grounded decision-making\n- Hallucination detection and mitigation in vision-language(-action) systems\n- Spatial, compositional, temporal, and relational reasoning in grounded multimodal systems\n- Causal reasoning, causal grounding, and counterfactual reasoning in multimodal systems\n- Training paradigms and architectures for grounded VLMs and VLAs\n- Reliable decision-making for embodied agents under dynamic and uncertain conditions\n- Out-of-distribution detection and failure mode analysis for grounded multimodal pipelines\n- Benchmarks, datasets, and evaluation methodologies for grounding and faithfulness\n- Interpretability and explainability of grounded multimodal reasoning\n- Emerging directions in grounded multimodal intelligence: world models and agentic VLMs\n\nImportant Dates\n- Paper Submission: Aug 30th, 2026\n- Notification: Sep 29th, 2026\n- Camera Ready: Oct, 2026\n- Workshop Date: Dec, 2026\n\nFAQ\nIs the workshop in-person or virtual? The workshop will be held in-person at NeurIPS 2026 in Sydney, Australia.\nWill the workshop proceedings be archival? No, the workshop proceedings will be non-archival. Authors of accepted papers retain the full copyright of their work and are free to submit extended versions to conferences or journals." }, { "key": "ICBINB-BIO", "title": "I Can't Believe It's Not Better (ICBINB): Failure Modes of AI in Biology", "subtitle": "ICBINB-BIO", "summary": "The biology branch of the I Can't Believe It's Not Better (ICBINB) initiative, this NeurIPS 2026 workshop stress-tests AI for biology in the real world - studying failure modes, robustness, and trustworthy scientific discovery by inviting negative results and rigorous evidence of where AI systems for biology fail.", "cfp_full": "I Can't Believe It's Not Better: Failure Modes of AI in Biology\nWorkshop at NeurIPS 2026 in Sydney\n\nStress-testing AI for biology in the real world: failure modes, robustness, and trustworthy scientific discovery.\n\nBenchmarks are only the beginning.\n\nAI is reshaping genomics, cellular modeling, structural biology, and therapeutic discovery. Yet strong benchmark results often fail to survive new mutations, perturbations, individuals, assays, or deployment settings.\n\nThis full-day workshop brings machine learning and life science researchers together to study those failures directly: what breaks, why it breaks, how we should evaluate it, and what more reliable scientific systems require.\n\nCall for Papers\n\nThe workshop invites submissions focusing on negative results and rigorous evidence of where AI systems for biology fail. Papers should document unexpected challenges in developing, evaluating, and deploying models on biological tasks and data.\n\nTopics\n- Out-of-distribution generalization and domain shift\n- Failure under weak or confounded supervision\n- Causal mechanisms versus spurious correlation\n- Uncertainty, calibration, and decision-aware reliability\n- Interpretability and trustworthy biological inference\n- Learning with limited data and distribution shift\n- Deployment-relevant evaluation beyond benchmarks\n- Limits of foundation, multimodal, and agentic models for biology\n- Causal intervention and experimental design\n\nSubmission Categories\n\nFull Papers - Up to 8 pages (excluding references/appendices). Must include: a problem statement with clearly specified biological task and metrics; a proposed approach with core mechanisms and hypotheses; an observed outcome with quantitative evidence; and a failure analysis investigating why the approach underperformed.\n\nTiny Papers - Up to 4 pages of main text, requiring a problem statement and evidence of negative/unexpected outcomes (full causal analysis optional).\n\nTechnical Requirements\n- Use the provided LaTeX template.\n- Submit via OpenReview (double-blind).\n- Disclose LLM usage in a brief paragraph.\n- Include ethics/reproducibility statements (optional, non-paginated).\n- Unlimited appendices (reviewer discretion).\n- First-time authors welcome.\n\nICBINB Initiative\n\nThis workshop forms one workshop in a series as part of the larger I Can't Believe It's Not Better (ICBINB) activities. ICBINB-BIO is the official biology branch of the broader ICBINB initiative, which shares a commitment to candid failure analysis, negative results, unexpectedly strong simple baselines, and benchmarks that support the claims made from them.\n\nKey Dates (All deadlines are 11:59 p.m. Anywhere on Earth and remain subject to final confirmation.)\n- Paper submission: August 29, 2026 (11:59 p.m. AoE) - Tentative\n- Review period: August 29 - September 21, 2026 - Tentative\n- Acceptance notification: September 29, 2026 - Tentative\n- Camera-ready & poster: October 20, 2026 - Tentative\n- In-person workshop: December 11 or 12, 2026, Sydney, Australia (exact date to be confirmed) - Tentative\n\nContact: icbinbbio@gmail.com", "cfp_status": "published", "topics": [ "Out-of-distribution generalization and domain shift", "Failure under weak or confounded supervision", "Causal mechanisms versus spurious correlation", "Uncertainty, calibration, and decision-aware reliability", "Interpretability and trustworthy biological inference", "Learning with limited data and distribution shift", "Deployment-relevant evaluation beyond benchmarks", "Limits of foundation, multimodal, and agentic models for biology", "Causal intervention and experimental design" ], "important_dates": [ { "label": "Paper submission (tentative)", "date": "2026-08-29" }, { "label": "Review period (tentative)", "date": "2026-08-29 to 2026-09-21" }, { "label": "Acceptance notification (tentative)", "date": "2026-09-29" }, { "label": "Camera-ready & poster (tentative)", "date": "2026-10-20" }, { "label": "In-person workshop (tentative)", "date": "December 11 or 12, 2026" } ], "organizers": [ "Maria Brbić (EPFL)", "Peter Koo (Cold Spring Harbor Laboratory)", "Bianca M. Dumitrascu (Columbia University)", "Su-In Lee (University of Washington)", "Siba Smarak Panigrahi (EPFL)", "Masayuki Nagai (Moon) (Cold Spring Harbor Laboratory)", "Ozgur Yilmaz Beker (Oz) (Columbia University)", "Soham Gadgil (University of Washington)" ], "speakers": [ "Hoifung Poon (Microsoft Research)", "Mihaela van der Schaar (University of Cambridge)" ], "host_url": "https://icbinb-bio.github.io/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/ICBINB-BIO", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/ICBINB-BIO", "location": "Sydney, Australia", "city": "Sydney", "workshop_date": "2026-12-10", "contact": "icbinbbio@gmail.com", "tracks": [ { "key": "ICBINB-BIO", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/ICBINB-BIO", "submission_dates_raw": "" } ], "group": "science", "group_label": "AI for Science", "corpus": "I Can't Believe It's Not Better (ICBINB): Failure Modes of AI in Biology ICBINB-BIO The biology branch of the I Can't Believe It's Not Better (ICBINB) initiative, this NeurIPS 2026 workshop stress-tests AI for biology in the real world - studying failure modes, robustness, and trustworthy scientific discovery by inviting negative results and rigorous evidence of where AI systems for biology fail. Out-of-distribution generalization and domain shift Failure under weak or confounded supervision Causal mechanisms versus spurious correlation Uncertainty, calibration, and decision-aware reliability Interpretability and trustworthy biological inference Learning with limited data and distribution shift Deployment-relevant evaluation beyond benchmarks Limits of foundation, multimodal, and agentic models for biology Causal intervention and experimental design I Can't Believe It's Not Better: Failure Modes of AI in Biology\nWorkshop at NeurIPS 2026 in Sydney\n\nStress-testing AI for biology in the real world: failure modes, robustness, and trustworthy scientific discovery.\n\nBenchmarks are only the beginning.\n\nAI is reshaping genomics, cellular modeling, structural biology, and therapeutic discovery. Yet strong benchmark results often fail to survive new mutations, perturbations, individuals, assays, or deployment settings.\n\nThis full-day workshop brings machine learning and life science researchers together to study those failures directly: what breaks, why it breaks, how we should evaluate it, and what more reliable scientific systems require.\n\nCall for Papers\n\nThe workshop invites submissions focusing on negative results and rigorous evidence of where AI systems for biology fail. Papers should document unexpected challenges in developing, evaluating, and deploying models on biological tasks and data.\n\nTopics\n- Out-of-distribution generalization and domain shift\n- Failure under weak or confounded supervision\n- Causal mechanisms versus spurious correlation\n- Uncertainty, calibration, and decision-aware reliability\n- Interpretability and trustworthy biological inference\n- Learning with limited data and distribution shift\n- Deployment-relevant evaluation beyond benchmarks\n- Limits of foundation, multimodal, and agentic models for biology\n- Causal intervention and experimental design\n\nSubmission Categories\n\nFull Papers - Up to 8 pages (excluding references/appendices). Must include: a problem statement with clearly specified biological task and metrics; a proposed approach with core mechanisms and hypotheses; an observed outcome with quantitative evidence; and a failure analysis investigating why the approach underperformed.\n\nTiny Papers - Up to 4 pages of main text, requiring a problem statement and evidence of negative/unexpected outcomes (full causal analysis optional).\n\nTechnical Requirements\n- Use the provided LaTeX template.\n- Submit via OpenReview (double-blind).\n- Disclose LLM usage in a brief paragraph.\n- Include ethics/reproducibility statements (optional, non-paginated).\n- Unlimited appendices (reviewer discretion).\n- First-time authors welcome.\n\nICBINB Initiative\n\nThis workshop forms one workshop in a series as part of the larger I Can't Believe It's Not Better (ICBINB) activities. ICBINB-BIO is the official biology branch of the broader ICBINB initiative, which shares a commitment to candid failure analysis, negative results, unexpectedly strong simple baselines, and benchmarks that support the claims made from them.\n\nKey Dates (All deadlines are 11:59 p.m. Anywhere on Earth and remain subject to final confirmation.)\n- Paper submission: August 29, 2026 (11:59 p.m. AoE) - Tentative\n- Review period: August 29 - September 21, 2026 - Tentative\n- Acceptance notification: September 29, 2026 - Tentative\n- Camera-ready & poster: October 20, 2026 - Tentative\n- In-person workshop: December 11 or 12, 2026, Sydney, Australia (exact date to be confirmed) - Tentative\n\nContact: icbinbbio@gmail.com" }, { "key": "InterpScience", "title": "Interpretability as a Science: NeurIPS 2026 Workshop", "subtitle": "InterpScience 2026", "summary": "A workshop asking what it would take to ground interpretability of large language models as a rigorous empirical science—addressing measurement, causality, and falsifiability, and drawing lessons from disciplines such as neuroscience, statistics, and causal representation learning.", "cfp_full": "Interpretability as a Science\nToward rigorous foundations for understanding LLMs\n\nAbout\nAs large language models grow in capability, interpretability asks how and why they behave as they do. Yet the field has not converged on notions of explanations at varying levels of abstraction, what evidence supports a claim, or how to design experiments that rule out alternative explanations. This workshop asks what it would take to ground interpretability as a rigorous empirical science—drawing lessons from disciplines that have long studied complex systems.\n\nWe focus on the following questions:\n- What does it mean to understand an LLM?\n- What standards, benchmarks, or evaluation criteria the field could adopt for measurement, causal claims, and falsifiability?\n- What can interpretability learn from neuroscience, statistics, and causal representation learning?\n\nA distinctive feature of this workshop is its interactive format. The workshop will host multiple breakout sessions, each led by a facilitator from a relevant discipline, who will give a short lightning talk highlighting important questions related to a specific sub-theme, then moderate a discussion connecting these ideas to interpretability. This format is designed to encourage genuine dialogue and engagement, inviting attendees to collectively shape a shared scientific foundation for interpretability.\n\nCall for Papers\nThe workshop invites submissions engaging with the scientific foundations of LLM interpretability—measurement, causality, falsifiability, and lessons from adjacent sciences.\n\nTopics of Interest\n- Understanding models and criteria for genuine explanation\n- Formal and mathematical frameworks for interpretability\n- Causal and interventional methods for grounding interpretability claims\n- Measurement validity, identifiability, and evaluation design\n- Falsifiability and experimental designs distinguishing mechanisms from artifacts\n- Transferable lessons from neuroscience, cognitive science, statistics, econometrics, and related fields\n- Pathways connecting interpretability findings to replicable science\n\nSubmission Guidelines\n- Accepts short papers (up to 5 pages) and long papers (up to 9 pages).\n- Submissions use either ICLR or NeurIPS formats; camera-ready must be NeurIPS format.\n- Accepted papers are non-archival, presented as posters, with some selected for talks.\n- At least one reciprocal reviewer required per submission.\n- Reviewers assess 2-3 papers during September 3-17, 2026.\n- Authors are responsible for verifying correctness and originality; fabricated citations result in desk rejection.\n\nDual Submission Policy\n- Previously published work at other ML conferences not accepted.\n- Work accepted to main NeurIPS conference cannot appear at workshop.\n- Concurrent NeurIPS submissions allowed but won't receive poster/presentation slots if main conference accepts them.\n- Submissions under review at other workshops not permitted.\n\nImportant Dates\n- Paper submission: August 28, 2026\n- Author notification: September 29, 2026\n- Camera-ready deadline: November 15, 2026\n- Workshop dates: December 11 or 12, 2026", "cfp_status": "published", "topics": [ "Understanding models and criteria for genuine explanation", "Formal and mathematical frameworks for interpretability", "Causal and interventional methods for grounding interpretability claims", "Measurement validity, identifiability, and evaluation design", "Falsifiability and experimental designs distinguishing mechanisms from artifacts", "Transferable lessons from neuroscience, cognitive science, statistics, econometrics, and related fields", "Pathways connecting interpretability findings to replicable science" ], "important_dates": [ { "label": "Paper submission", "date": "2026-08-28" }, { "label": "Author notification", "date": "2026-09-29" }, { "label": "Camera-ready deadline", "date": "2026-11-15" }, { "label": "Workshop dates", "date": "December 11 or 12, 2026" } ], "organizers": [ "Dhanya Sridhar (Mila & Université de Montréal)", "Navita Goyal (University of Maryland)", "Patrik Reizinger (MPI-IS Tübingen)", "Shruti Joshi (Mila & Université de Montréal)", "Gemma Moran (Rutgers University)", "David Klindt (Cold Spring Harbor Laboratory)", "Hal Daumé III (University of Maryland)" ], "speakers": [], "host_url": "https://interpscience.github.io/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/InterpScience", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/InterpScience", "location": "Sydney, Australia", "city": "Sydney", "workshop_date": "2026-12-10", "contact": "interpscience@gmail.com", "tracks": [ { "key": "InterpScience", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/InterpScience", "submission_dates_raw": "" } ], "group": "trust", "group_label": "Trustworthy & Safe AI", "corpus": "Interpretability as a Science: NeurIPS 2026 Workshop InterpScience 2026 A workshop asking what it would take to ground interpretability of large language models as a rigorous empirical science—addressing measurement, causality, and falsifiability, and drawing lessons from disciplines such as neuroscience, statistics, and causal representation learning. Understanding models and criteria for genuine explanation Formal and mathematical frameworks for interpretability Causal and interventional methods for grounding interpretability claims Measurement validity, identifiability, and evaluation design Falsifiability and experimental designs distinguishing mechanisms from artifacts Transferable lessons from neuroscience, cognitive science, statistics, econometrics, and related fields Pathways connecting interpretability findings to replicable science Interpretability as a Science\nToward rigorous foundations for understanding LLMs\n\nAbout\nAs large language models grow in capability, interpretability asks how and why they behave as they do. Yet the field has not converged on notions of explanations at varying levels of abstraction, what evidence supports a claim, or how to design experiments that rule out alternative explanations. This workshop asks what it would take to ground interpretability as a rigorous empirical science—drawing lessons from disciplines that have long studied complex systems.\n\nWe focus on the following questions:\n- What does it mean to understand an LLM?\n- What standards, benchmarks, or evaluation criteria the field could adopt for measurement, causal claims, and falsifiability?\n- What can interpretability learn from neuroscience, statistics, and causal representation learning?\n\nA distinctive feature of this workshop is its interactive format. The workshop will host multiple breakout sessions, each led by a facilitator from a relevant discipline, who will give a short lightning talk highlighting important questions related to a specific sub-theme, then moderate a discussion connecting these ideas to interpretability. This format is designed to encourage genuine dialogue and engagement, inviting attendees to collectively shape a shared scientific foundation for interpretability.\n\nCall for Papers\nThe workshop invites submissions engaging with the scientific foundations of LLM interpretability—measurement, causality, falsifiability, and lessons from adjacent sciences.\n\nTopics of Interest\n- Understanding models and criteria for genuine explanation\n- Formal and mathematical frameworks for interpretability\n- Causal and interventional methods for grounding interpretability claims\n- Measurement validity, identifiability, and evaluation design\n- Falsifiability and experimental designs distinguishing mechanisms from artifacts\n- Transferable lessons from neuroscience, cognitive science, statistics, econometrics, and related fields\n- Pathways connecting interpretability findings to replicable science\n\nSubmission Guidelines\n- Accepts short papers (up to 5 pages) and long papers (up to 9 pages).\n- Submissions use either ICLR or NeurIPS formats; camera-ready must be NeurIPS format.\n- Accepted papers are non-archival, presented as posters, with some selected for talks.\n- At least one reciprocal reviewer required per submission.\n- Reviewers assess 2-3 papers during September 3-17, 2026.\n- Authors are responsible for verifying correctness and originality; fabricated citations result in desk rejection.\n\nDual Submission Policy\n- Previously published work at other ML conferences not accepted.\n- Work accepted to main NeurIPS conference cannot appear at workshop.\n- Concurrent NeurIPS submissions allowed but won't receive poster/presentation slots if main conference accepts them.\n- Submissions under review at other workshops not permitted.\n\nImportant Dates\n- Paper submission: August 28, 2026\n- Author notification: September 29, 2026\n- Camera-ready deadline: November 15, 2026\n- Workshop dates: December 11 or 12, 2026" }, { "key": "LXAI", "title": "LatinX in AI Workshop @ NeurIPS 2026", "subtitle": "LXAI @ NeurIPS 2026", "summary": "The LatinX in AI (LXAI) Research Workshop at NeurIPS 2026 is a one-day event with invited speakers, oral presentations, and posters that showcases research by the LatinX community across all NeurIPS subject areas and provides mentoring and networking.", "cfp_full": "LatinX in AI Research Workshop at NeurIPS 2026\nDecember, 2026 | Sidney, Australia & Atlanta, USA\n\nThis is an official workshop of the LatinX in AI (LXAI) organization at NeurIPS.\n\nThe workshop is a one-day event with invited speakers, oral presentations, and posters. The event brings together faculty, graduate students, research scientists, and engineers for an opportunity to connect and exchange ideas. There will be a panel discussion and a mentoring session to discuss current research trends and career choices in artificial intelligence and machine learning. While all presenters will identify primarily as LatinX, everyone is invited to attend. If you have any questions feel free to contact the workshop chairs at: neurips-2026@latinxinai.org\n\nIMPORTANT DATES\n\nCall for Papers --- July 22nd, 2026\n- Submission Open: July 22nd, 2026\n- Submission deadline: August 20th, 2026\n- Notification of acceptance: September 2nd, 2026\n\nCall for reviewers -- July 31st, 2026\n- Submission Open: July 22nd, 2026\n- Submission deadline: August 18th, 2026\n- Review Allocation: August 20th, 2026\n- Review Deadline: August 29th, 2026\n\nLXAI Workshop --- TBD\n\n*All deadlines are 23:59:59 AoE (UTC-12)\n\nCall For Papers\n\nLXAI will receive papers in any subject accepted by NeurIPS. At least one of the authors must identify as LatinX. Submissions will be double-blind peer-reviewed (that is, make sure to not reveal your names, institutions, or country anywhere in the document) and should be submitted as a PDF file through OpenReview. All submissions must be in English and strictly follow the format provided by the NeurIPS 2026. Papers using any other format will be desk rejected without review. You can submit your paper to one of two categories:\n- Extended Abstract: Up to 4 pages, excluding references.\n- Full Paper: Up to 8 pages long, excluding references. (papers with less than 6 pages might be rejected without review or automatically delegated to the extended abstract category).\nThe number of pages above includes space taken by figures, tables, and equations. At the authors' discretion.\n\nSpecific NeurIPS topics include, but are not limited to:\n- General Machine Learning\n- Deep Learning\n- Evaluation (methodology, meta studies, replicability and validity, human-in-the-loop, etc.)\n- Theory of Machine Learning\n- Machine Learning Systems\n- Optimization\n- Probabilistic Methods\n- Reinforcement Learning\n- Trustworthy Machine Learning\n- Application-Driven Machine Learning\n\nPresentation Location\nWe are still coordinating with the main NeurIPS organization regarding the split of the conference between Sidney and Atlanta. In the submission form you have the opportunity to mark visa restrictions and location preferences, we will communicate upon acceptance if you will be required to present in Sidney or Atlanta. For the Paris co-location, authors can participate at the Affinity Groups Joint Poster session only.\n\nPublication Options\nAccepted papers will be invited for inclusion in the Journal of LatinX in AI Research (JLXAIR). Authors may choose between:\n- Archival Publication: Paper permanently published in JLXAIR; Included in official journal records; Receives DOI and formal citation; Cannot be submitted elsewhere after acceptance.\n- Non-Archival Publication: Presented at LXAI workshop; Not included in permanent journal archive; Authors retain right to submit to other venues; Ideal for work also submitted to main conference.\nAuthors must indicate their publication preference upon paper acceptance.\n\nCommitment to Diversity and Inclusion\nLXAI is committed to creating an inclusive environment that welcomes researchers from all backgrounds. We encourage submissions from intersectional identities and value diverse perspectives in advancing AI research.\n\nWe look forward to your submissions and to celebrating the outstanding contributions of the LatinX AI community at NeurIPS 2026!", "cfp_status": "published", "topics": [ "General Machine Learning", "Deep Learning", "Evaluation (methodology, meta studies, replicability and validity, human-in-the-loop, etc.)", "Theory of Machine Learning", "Machine Learning Systems", "Optimization", "Probabilistic Methods", "Reinforcement Learning", "Trustworthy Machine Learning", "Application-Driven Machine Learning" ], "important_dates": [ { "label": "Submission Open", "date": "2026-07-22" }, { "label": "Submission deadline", "date": "2026-08-20" }, { "label": "Notification of acceptance", "date": "2026-09-02" }, { "label": "Review Allocation", "date": "2026-08-20" }, { "label": "Review Deadline", "date": "2026-08-29" } ], "organizers": [], "speakers": [], "host_url": "https://www.latinxinai.org/neurips-2026", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/LXAI", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/LXAI", "location": "Sidney, Australia", "city": "Sydney", "workshop_date": "2026-07-22", "contact": "neurips-2026@latinxinai.org", "tracks": [ { "key": "LXAI", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/LXAI", "submission_dates_raw": "" } ], "group": "society", "group_label": "Society, Applications & Community", "corpus": "LatinX in AI Workshop @ NeurIPS 2026 LXAI @ NeurIPS 2026 The LatinX in AI (LXAI) Research Workshop at NeurIPS 2026 is a one-day event with invited speakers, oral presentations, and posters that showcases research by the LatinX community across all NeurIPS subject areas and provides mentoring and networking. General Machine Learning Deep Learning Evaluation (methodology, meta studies, replicability and validity, human-in-the-loop, etc.) Theory of Machine Learning Machine Learning Systems Optimization Probabilistic Methods Reinforcement Learning Trustworthy Machine Learning Application-Driven Machine Learning LatinX in AI Research Workshop at NeurIPS 2026\nDecember, 2026 | Sidney, Australia & Atlanta, USA\n\nThis is an official workshop of the LatinX in AI (LXAI) organization at NeurIPS.\n\nThe workshop is a one-day event with invited speakers, oral presentations, and posters. The event brings together faculty, graduate students, research scientists, and engineers for an opportunity to connect and exchange ideas. There will be a panel discussion and a mentoring session to discuss current research trends and career choices in artificial intelligence and machine learning. While all presenters will identify primarily as LatinX, everyone is invited to attend. If you have any questions feel free to contact the workshop chairs at: neurips-2026@latinxinai.org\n\nIMPORTANT DATES\n\nCall for Papers --- July 22nd, 2026\n- Submission Open: July 22nd, 2026\n- Submission deadline: August 20th, 2026\n- Notification of acceptance: September 2nd, 2026\n\nCall for reviewers -- July 31st, 2026\n- Submission Open: July 22nd, 2026\n- Submission deadline: August 18th, 2026\n- Review Allocation: August 20th, 2026\n- Review Deadline: August 29th, 2026\n\nLXAI Workshop --- TBD\n\n*All deadlines are 23:59:59 AoE (UTC-12)\n\nCall For Papers\n\nLXAI will receive papers in any subject accepted by NeurIPS. At least one of the authors must identify as LatinX. Submissions will be double-blind peer-reviewed (that is, make sure to not reveal your names, institutions, or country anywhere in the document) and should be submitted as a PDF file through OpenReview. All submissions must be in English and strictly follow the format provided by the NeurIPS 2026. Papers using any other format will be desk rejected without review. You can submit your paper to one of two categories:\n- Extended Abstract: Up to 4 pages, excluding references.\n- Full Paper: Up to 8 pages long, excluding references. (papers with less than 6 pages might be rejected without review or automatically delegated to the extended abstract category).\nThe number of pages above includes space taken by figures, tables, and equations. At the authors' discretion.\n\nSpecific NeurIPS topics include, but are not limited to:\n- General Machine Learning\n- Deep Learning\n- Evaluation (methodology, meta studies, replicability and validity, human-in-the-loop, etc.)\n- Theory of Machine Learning\n- Machine Learning Systems\n- Optimization\n- Probabilistic Methods\n- Reinforcement Learning\n- Trustworthy Machine Learning\n- Application-Driven Machine Learning\n\nPresentation Location\nWe are still coordinating with the main NeurIPS organization regarding the split of the conference between Sidney and Atlanta. In the submission form you have the opportunity to mark visa restrictions and location preferences, we will communicate upon acceptance if you will be required to present in Sidney or Atlanta. For the Paris co-location, authors can participate at the Affinity Groups Joint Poster session only.\n\nPublication Options\nAccepted papers will be invited for inclusion in the Journal of LatinX in AI Research (JLXAIR). Authors may choose between:\n- Archival Publication: Paper permanently published in JLXAIR; Included in official journal records; Receives DOI and formal citation; Cannot be submitted elsewhere after acceptance.\n- Non-Archival Publication: Presented at LXAI workshop; Not included in permanent journal archive; Authors retain right to submit to other venues; Ideal for work also submitted to main conference.\nAuthors must indicate their publication preference upon paper acceptance.\n\nCommitment to Diversity and Inclusion\nLXAI is committed to creating an inclusive environment that welcomes researchers from all backgrounds. We encourage submissions from intersectional identities and value diverse perspectives in advancing AI research.\n\nWe look forward to your submissions and to celebrating the outstanding contributions of the LatinX AI community at NeurIPS 2026!" }, { "key": "MLForSys", "title": "Machine Learning for Systems 2026", "subtitle": "MLForSys2026", "summary": "The Workshop on ML for Systems at NeurIPS 2026 presents cutting-edge work applying machine learning to computer systems problems, and aims to develop a unified methodology for the field while exploring how LLMs, multimodal foundation models, and agentic workflows can address systems challenges.", "cfp_full": "Workshop on ML for Systems at NeurIPS 2026, December 11 or 12 (TBA), International Convention Center Sydney\nWorkshop on ML for Systems at NeurIPS '26, Sydney, Australia\n\nWhat To Expect\n\nThe ML for Systems workshop presents cutting-edge work on ML in computer systems and aims to develop a unified methodology for the field.\n\nMachine Learning (ML) for Systems describes the application of machine learning techniques to problems related to computer systems. By leveraging supervised learning and reinforcement learning (RL) approaches, machine learning can replace longstanding heuristics that currently drive many of these systems. This includes a wide range of topics, including multi-objective tasks such as designing new data structures, integrated circuits, or design verification, as well as implementing control algorithms for applications such as compilers, databases, memory management, or ML frameworks. While the systems community increasingly recognizes the importance of ML in solving a variety of different systems problems, ML for Systems remains an emerging area without widely established best practices, methods and strategies for the application of state-of-the-art machine learning techniques. The goal of this workshop is to provide an interdisciplinary venue for ML and Systems experts to push this boundary and start new directions within the ML for Systems area.\n\nWorkshop Direction\n\nIn previous 9 editions, we showcased specific approaches and frameworks to solve problems, bringing together researchers and practitioners at NeurIPS from both the ML and systems communities. While breaking new grounds, we encouraged collaborations and development in a broad range of ML for Systems works, many later published in top-tier conferences. This year, we plan to continue this path while exploring how emerging ML paradigms—including large language models (LLMs), multimodal foundation models, and agentic workflows—can be leveraged to address systems challenges and improve the efficiency, reliability, and scalability of ML infrastructure itself.\n\nRecently, the rise of LLMs, multimodal foundation models, and agentic workflows has presented new opportunities and challenges within the domain of computer systems. Our community is well-positioned to produce science and stimulate discussion for adapting to this new paradigm. We seek to explore both how these models can be used to solve systems problems, and how to address systems issues that emerge from large-scale training and serving of such models. Additionally, we place emphasis on developing best practices, methodologies, benchmarks, datasets, simulators, and evaluation frameworks that improve rigor, reproducibility, reliability, and trustworthiness in ML for Systems research.\n\nWorkshop Goals\n\nNeurIPS provides a unique opportunity to bring together systems researchers and researchers from other sub-areas of ML who had not previously considered applying their techniques in a computer systems context. We see the goal of our workshop as solving the following two objectives:\n- Opening up connections between research areas that were not previously considered, connecting the ML and Systems communities, growing the scope of ML for Systems work and unlocking new research opportunities.\n- Developing best practices, methodologies and benchmarks for the ML for Systems field.\n\nOur program will include a variety of speakers and poster sessions from selected papers. We invite researchers to submit relevant papers through our call for papers.\n\nSubmission Instructions (from the Call for Papers)\nSubmissions should be up to 4-page extended abstracts (with optional additional appendix material), in PDF format, following the NeurIPS 2026 formatting guidelines. No anonymization is required. Papers are submitted via OpenReview at the designated workshop portal. Topics of interest include LLMs and agentic workflows for systems problems (hardware design, compiler optimization, debugging, design-space exploration); ML for AI infrastructure challenges including distributed training, resource allocation, and inference optimization; methodologies and frameworks for reproducibility, benchmarking, and trustworthiness; and applications spanning systems software, distributed systems, compilers, databases, computer architecture, networking, storage, data centers, sustainable computing, and LLM systems reliability.\n\nImportant Dates\n- Submission Deadline: August 29, 2026 (midnight, Anywhere on Earth)\n- Acceptance Notifications: September 29, 2026\n- Workshop Date: December 11 or 12, 2026 (details TBA)", "cfp_status": "published", "topics": [ "LLMs and agentic workflows for systems problems (hardware design, compiler optimization, debugging, design-space exploration)", "ML for AI infrastructure (distributed training, resource allocation, inference optimization)", "Applying supervised and reinforcement learning to replace systems heuristics", "Designing data structures, integrated circuits, and design verification", "Control algorithms for compilers, databases, and memory management", "Best practices, methodologies, benchmarks, datasets, simulators, and evaluation frameworks", "Systems issues from large-scale training and serving of foundation models", "Reproducibility, reliability, and trustworthiness in ML for Systems research" ], "important_dates": [ { "label": "Submission Deadline", "date": "2026-08-29" }, { "label": "Acceptance Notifications", "date": "2026-09-29" }, { "label": "Workshop Date", "date": "2026-12-11 or 2026-12-12" } ], "organizers": [ "Divya Mahajan (Georgia Tech)", "Patrick Musau (Google)", "Phitchaya Mangpo Phothilimthana (OpenAI)", "Haoran Qiu (Microsoft Azure Research)", "Mimee Xu (NYU)", "Dan Zhang (Recursive Intelligence)" ], "speakers": [], "host_url": "https://mlforsystems.org", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/MLForSys", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/MLForSys", "location": "Sydney, Australia", "city": "Sydney", "workshop_date": "2026-12-11", "contact": "divya.mahajan@gatech.edu", "tracks": [ { "key": "MLForSys", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/MLForSys", "submission_dates_raw": "" } ], "group": "society", "group_label": "Society, Applications & Community", "corpus": "Machine Learning for Systems 2026 MLForSys2026 The Workshop on ML for Systems at NeurIPS 2026 presents cutting-edge work applying machine learning to computer systems problems, and aims to develop a unified methodology for the field while exploring how LLMs, multimodal foundation models, and agentic workflows can address systems challenges. LLMs and agentic workflows for systems problems (hardware design, compiler optimization, debugging, design-space exploration) ML for AI infrastructure (distributed training, resource allocation, inference optimization) Applying supervised and reinforcement learning to replace systems heuristics Designing data structures, integrated circuits, and design verification Control algorithms for compilers, databases, and memory management Best practices, methodologies, benchmarks, datasets, simulators, and evaluation frameworks Systems issues from large-scale training and serving of foundation models Reproducibility, reliability, and trustworthiness in ML for Systems research Workshop on ML for Systems at NeurIPS 2026, December 11 or 12 (TBA), International Convention Center Sydney\nWorkshop on ML for Systems at NeurIPS '26, Sydney, Australia\n\nWhat To Expect\n\nThe ML for Systems workshop presents cutting-edge work on ML in computer systems and aims to develop a unified methodology for the field.\n\nMachine Learning (ML) for Systems describes the application of machine learning techniques to problems related to computer systems. By leveraging supervised learning and reinforcement learning (RL) approaches, machine learning can replace longstanding heuristics that currently drive many of these systems. This includes a wide range of topics, including multi-objective tasks such as designing new data structures, integrated circuits, or design verification, as well as implementing control algorithms for applications such as compilers, databases, memory management, or ML frameworks. While the systems community increasingly recognizes the importance of ML in solving a variety of different systems problems, ML for Systems remains an emerging area without widely established best practices, methods and strategies for the application of state-of-the-art machine learning techniques. The goal of this workshop is to provide an interdisciplinary venue for ML and Systems experts to push this boundary and start new directions within the ML for Systems area.\n\nWorkshop Direction\n\nIn previous 9 editions, we showcased specific approaches and frameworks to solve problems, bringing together researchers and practitioners at NeurIPS from both the ML and systems communities. While breaking new grounds, we encouraged collaborations and development in a broad range of ML for Systems works, many later published in top-tier conferences. This year, we plan to continue this path while exploring how emerging ML paradigms—including large language models (LLMs), multimodal foundation models, and agentic workflows—can be leveraged to address systems challenges and improve the efficiency, reliability, and scalability of ML infrastructure itself.\n\nRecently, the rise of LLMs, multimodal foundation models, and agentic workflows has presented new opportunities and challenges within the domain of computer systems. Our community is well-positioned to produce science and stimulate discussion for adapting to this new paradigm. We seek to explore both how these models can be used to solve systems problems, and how to address systems issues that emerge from large-scale training and serving of such models. Additionally, we place emphasis on developing best practices, methodologies, benchmarks, datasets, simulators, and evaluation frameworks that improve rigor, reproducibility, reliability, and trustworthiness in ML for Systems research.\n\nWorkshop Goals\n\nNeurIPS provides a unique opportunity to bring together systems researchers and researchers from other sub-areas of ML who had not previously considered applying their techniques in a computer systems context. We see the goal of our workshop as solving the following two objectives:\n- Opening up connections between research areas that were not previously considered, connecting the ML and Systems communities, growing the scope of ML for Systems work and unlocking new research opportunities.\n- Developing best practices, methodologies and benchmarks for the ML for Systems field.\n\nOur program will include a variety of speakers and poster sessions from selected papers. We invite researchers to submit relevant papers through our call for papers.\n\nSubmission Instructions (from the Call for Papers)\nSubmissions should be up to 4-page extended abstracts (with optional additional appendix material), in PDF format, following the NeurIPS 2026 formatting guidelines. No anonymization is required. Papers are submitted via OpenReview at the designated workshop portal. Topics of interest include LLMs and agentic workflows for systems problems (hardware design, compiler optimization, debugging, design-space exploration); ML for AI infrastructure challenges including distributed training, resource allocation, and inference optimization; methodologies and frameworks for reproducibility, benchmarking, and trustworthiness; and applications spanning systems software, distributed systems, compilers, databases, computer architecture, networking, storage, data centers, sustainable computing, and LLM systems reliability.\n\nImportant Dates\n- Submission Deadline: August 29, 2026 (midnight, Anywhere on Earth)\n- Acceptance Notifications: September 29, 2026\n- Workshop Date: December 11 or 12, 2026 (details TBA)" }, { "key": "NeuralArtifacts", "title": "Neural Network Artifacts as a New Data Modality", "subtitle": "NeuralArtifacts", "summary": "A workshop treating neural network artifacts (weights, gradients, intermediate representations, optimization trajectories, and other computational traces) as a data modality in their own right, connecting communities working on model merging, meta-learning, mechanistic interpretability, neural architecture search, and neural fields under a shared data-centric perspective.", "cfp_full": "Overview\n\nMachine learning has revolutionized how we learn from scientific data, yet it has rarely turned that same population-level lens on its own products. This workshop aims to close that gap by treating neural network artifacts as a data modality in their own right.\n\nToday's model repositories contain immense distributed knowledge encoded not only in neural network weights, but also in gradients, intermediate representations, optimization trajectories, and other computational traces. We refer to these collectively as neural artifacts. Learning from populations of these artifacts can help us compare, search, explain, modify, control, and synthesize models.\n\nFollowing the inaugural ICLR 2025 workshop, this second edition broadens the scope beyond weights and places greater emphasis on standardized datasets, benchmarks, tasks, neural lineages, and AI supply chains. Our goal is to connect communities working on model merging, meta-learning, mechanistic interpretability, neural architecture search, neural fields, and related areas under a shared data-centric perspective.\n\nWorkshop Themes\n\n- Datasets and benchmarks: Standardized model zoos, evaluation protocols, and new tasks for learning from neural artifacts.\n- Foundations and theory: Structure, symmetries, scaling laws, and specialized architectures such as equivariant metanetworks.\n- Model analysis and dynamics: Inferring behavior, generalization, safety, robustness, fairness, memorization, and backdoors from artifacts; understanding learning dynamics and interpretability.\n- Model synthesis and control: Generating and editing models through hypernetworks, task arithmetic, merging, steering, and related techniques.\n- Model search and selection: Navigating model populations to select models for inference, fine-tuning, or transfer without expensive retraining or evaluation.\n- Model populations and AI supply chains: Mapping model lineages, trends, knowledge gaps, and supply-chain effects through tools such as model atlases.\n\nResearch Goals and Key Questions\n\nThis workshop will explore questions such as:\n- How should neural artifacts be represented, compared, and modeled across architectures and training runs?\n- What can weights and computational traces reveal about model behavior, provenance, safety, and learning dynamics?\n- How can populations of models support efficient search, selection, transfer, merging, editing, and generation?\n- Which datasets, benchmarks, and evaluation protocols are needed to make progress measurable and reproducible?\n- How can insights from theory, interpretability, neural fields, and AI supply chains strengthen one another?\n\nCall for Papers\n\nThe workshop invites contributions treating neural network artifacts - including weights, gradients, optimization trajectories, internal representations, and other computational traces - as a learning data modality.\n\nTopics\n\n- Datasets and Benchmarks: Model zoos, neural-artifact datasets, evaluation protocols, and infrastructure for sharing model populations.\n- Foundations and Theory: Structure and scaling laws of neural artifacts, theoretical frameworks for weight-space learning, specialized architectures, and expressivity analysis.\n- Model Analysis and Dynamics: Performance prediction, optimization trajectories, interpretability through weights and gradients, and neural lineage tracking.\n- Model Synthesis and Control: Hypernetworks, model merging, task arithmetic, model editing, and safety interventions.\n- Model Search and Selection: Navigating model populations and predicting model compatibility.\n- Model Populations and AI Supply Chains: Mapping model ecosystems and analyzing supply chain effects on weights.\n\nSubmission Tracks\n\n- Extended abstracts (4-6 pages, non-archival): Early-stage results, position papers, ideas, negative results, and benchmark proposals.\n- Full papers (8-12 pages, archival): Substantiated research contributions.\n\nPage limits exclude references and supplementary material.\n\nSubmission Instructions\n\nSubmissions use OpenReview. Work must follow the NeurIPS 2026 formatting and generative-AI guidelines and be self-contained.\n\nImportant Dates\n\n- Paper submission deadline: September 1, 2026 (Anywhere on Earth)\n- Other dates: To be announced", "cfp_status": "published", "topics": [ "Datasets and benchmarks: standardized model zoos, evaluation protocols, and new tasks for learning from neural artifacts", "Foundations and theory: structure, symmetries, scaling laws, and specialized architectures such as equivariant metanetworks", "Model analysis and dynamics: inferring behavior, generalization, safety, robustness, fairness, memorization, and backdoors from artifacts", "Model synthesis and control: hypernetworks, task arithmetic, merging, steering, and model editing", "Model search and selection: navigating model populations for inference, fine-tuning, or transfer", "Model populations and AI supply chains: mapping model lineages, trends, knowledge gaps, and supply-chain effects" ], "important_dates": [ { "label": "Paper submission deadline", "date": "2026-09-01" } ], "organizers": [], "speakers": [], "host_url": "https://artifactsasdata.org", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/NeuralArtifacts", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/NeuralArtifacts", "location": "Paris, France", "city": "Paris", "workshop_date": "2026-12-11", "contact": "organizers@artifactsasdata.org", "tracks": [ { "key": "NeuralArtifacts", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/NeuralArtifacts", "submission_dates_raw": "" } ], "group": "society", "group_label": "Society, Applications & Community", "corpus": "Neural Network Artifacts as a New Data Modality NeuralArtifacts A workshop treating neural network artifacts (weights, gradients, intermediate representations, optimization trajectories, and other computational traces) as a data modality in their own right, connecting communities working on model merging, meta-learning, mechanistic interpretability, neural architecture search, and neural fields under a shared data-centric perspective. Datasets and benchmarks: standardized model zoos, evaluation protocols, and new tasks for learning from neural artifacts Foundations and theory: structure, symmetries, scaling laws, and specialized architectures such as equivariant metanetworks Model analysis and dynamics: inferring behavior, generalization, safety, robustness, fairness, memorization, and backdoors from artifacts Model synthesis and control: hypernetworks, task arithmetic, merging, steering, and model editing Model search and selection: navigating model populations for inference, fine-tuning, or transfer Model populations and AI supply chains: mapping model lineages, trends, knowledge gaps, and supply-chain effects Overview\n\nMachine learning has revolutionized how we learn from scientific data, yet it has rarely turned that same population-level lens on its own products. This workshop aims to close that gap by treating neural network artifacts as a data modality in their own right.\n\nToday's model repositories contain immense distributed knowledge encoded not only in neural network weights, but also in gradients, intermediate representations, optimization trajectories, and other computational traces. We refer to these collectively as neural artifacts. Learning from populations of these artifacts can help us compare, search, explain, modify, control, and synthesize models.\n\nFollowing the inaugural ICLR 2025 workshop, this second edition broadens the scope beyond weights and places greater emphasis on standardized datasets, benchmarks, tasks, neural lineages, and AI supply chains. Our goal is to connect communities working on model merging, meta-learning, mechanistic interpretability, neural architecture search, neural fields, and related areas under a shared data-centric perspective.\n\nWorkshop Themes\n\n- Datasets and benchmarks: Standardized model zoos, evaluation protocols, and new tasks for learning from neural artifacts.\n- Foundations and theory: Structure, symmetries, scaling laws, and specialized architectures such as equivariant metanetworks.\n- Model analysis and dynamics: Inferring behavior, generalization, safety, robustness, fairness, memorization, and backdoors from artifacts; understanding learning dynamics and interpretability.\n- Model synthesis and control: Generating and editing models through hypernetworks, task arithmetic, merging, steering, and related techniques.\n- Model search and selection: Navigating model populations to select models for inference, fine-tuning, or transfer without expensive retraining or evaluation.\n- Model populations and AI supply chains: Mapping model lineages, trends, knowledge gaps, and supply-chain effects through tools such as model atlases.\n\nResearch Goals and Key Questions\n\nThis workshop will explore questions such as:\n- How should neural artifacts be represented, compared, and modeled across architectures and training runs?\n- What can weights and computational traces reveal about model behavior, provenance, safety, and learning dynamics?\n- How can populations of models support efficient search, selection, transfer, merging, editing, and generation?\n- Which datasets, benchmarks, and evaluation protocols are needed to make progress measurable and reproducible?\n- How can insights from theory, interpretability, neural fields, and AI supply chains strengthen one another?\n\nCall for Papers\n\nThe workshop invites contributions treating neural network artifacts - including weights, gradients, optimization trajectories, internal representations, and other computational traces - as a learning data modality.\n\nTopics\n\n- Datasets and Benchmarks: Model zoos, neural-artifact datasets, evaluation protocols, and infrastructure for sharing model populations.\n- Foundations and Theory: Structure and scaling laws of neural artifacts, theoretical frameworks for weight-space learning, specialized architectures, and expressivity analysis.\n- Model Analysis and Dynamics: Performance prediction, optimization trajectories, interpretability through weights and gradients, and neural lineage tracking.\n- Model Synthesis and Control: Hypernetworks, model merging, task arithmetic, model editing, and safety interventions.\n- Model Search and Selection: Navigating model populations and predicting model compatibility.\n- Model Populations and AI Supply Chains: Mapping model ecosystems and analyzing supply chain effects on weights.\n\nSubmission Tracks\n\n- Extended abstracts (4-6 pages, non-archival): Early-stage results, position papers, ideas, negative results, and benchmark proposals.\n- Full papers (8-12 pages, archival): Substantiated research contributions.\n\nPage limits exclude references and supplementary material.\n\nSubmission Instructions\n\nSubmissions use OpenReview. Work must follow the NeurIPS 2026 formatting and generative-AI guidelines and be self-contained.\n\nImportant Dates\n\n- Paper submission deadline: September 1, 2026 (Anywhere on Earth)\n- Other dates: To be announced" }, { "key": "AgenticLS", "title": "NeurIPS 2026 Agentic AI for Biological Discovery Workshop", "subtitle": "NeurIPS 2026 Workshop AgenticLS", "summary": "A NeurIPS 2026 workshop on building, benchmarking, and deploying agentic AI systems for life-science discovery — from scientific copilots to closed-loop laboratories — spanning agent design, generalist-vs-specialist model choices, and autonomous lab-in-the-loop experimentation.", "cfp_full": "Agentic AI for Biological Discovery (AgenticLS)\nFrom scientific copilots to closed-loop labs: building, benchmarking, and deploying agentic AI systems for the next generation of life-science discovery.\n\nAbout the workshop\nThe life sciences are entering an era of agentic AI — systems built on tool-using and reasoning frameworks that go beyond static prediction to read literature, call specialized tools, plan multi-step analyses, propose experiments, and in some cases interact directly with laboratories and robotics. This shift is enabled both by frontier LLMs and by a rapidly growing stack of biology-specialized foundation models for proteins, genomes, and single cells.\n\nYet the field remains strikingly young. There is still little consensus on how to build effective life-science agents, when biology-specialized models are necessary versus when general-purpose LLMs suffice, and how to deploy such agents toward the ultimate goal: accelerating biological discovery and drug development. AgenticLS brings together researchers from machine learning, computational biology, experimental biology, drug discovery, and lab automation to tackle these questions as a building, deployment, and evaluation problem.\n\nCore themes\n1. Building agents for discovery: How should we design agent harnesses, orchestrate multi-agent systems, manage long-horizon memory and context, and integrate biological tool ecosystems? What metrics and rewards guide agents toward novel discoveries?\n2. Generalist vs. specialist: When are frontier general-purpose models (e.g., GPT, Claude) sufficient, and when do biology-specialized models (e.g., AlphaFold, ESM) materially improve planning, reasoning, or downstream outcomes — and can specialists improve via recursive self-improvement?\n3. Autonomous lab-in-the-loop: How should agentic AI autonomously interface with robotics, assay platforms, and human scientists to enable reliable, iterative experimentation in closed-loop laboratories?\n\nScope & topics\nWe welcome submissions across two tracks spanning the full agentic life-science stack.\n\nTrack I — Building Agentic Systems for Life Science\n- Agent architectures & reasoning: LLM-based, multimodal, and retrieval-augmented agents; long-horizon planners; multi-agent scientific collaborations.\n- Harness design & orchestration: tool ecosystems and APIs for biology, multi-agent communication and routing, long-horizon memory and context management, knowledge grounding.\n- Agentic literature & knowledge systems: hypothesis generation, evidence synthesis, claim verification, protocol assistance, ontology-aware and knowledge-graph-enhanced agents.\n- Generalist vs. specialist comparisons: foundation-model backbones, bio-specific instruction tuning, domain adaptation, biological foundation models inside agent pipelines.\n- Reinforcement learning for scientific agents: RL from lab-in-the-loop feedback and reward-function design for discovery.\n- Recursive self-improvement for bio-specialized models: data generation and curation, automated evaluation, feedback-driven refinement, model-in-the-loop workflows.\n\nTrack II — Closed-Loop Discovery & Applications\n- Lab-in-the-loop & robotics: closed-loop experimentation, robotic execution, active learning, adaptive experiment planning, autonomous laboratories that learn from wet-lab feedback.\n- Agents for biological design & discovery: target identification, sequence and molecular design, structure-based design, CRISPR / perturbation design, assay planning, therapeutic optimization.\n- Evaluation & benchmarks: faithfulness, reproducibility, calibrated uncertainty, expert judgment, simulator-to-lab transfer, biological validity, system-level benchmarks.\n- Reliability, governance & biosafety: hallucinated biology, biosecurity-aware safeguards, human oversight, traceability, auditability, and deployment standards for scientific agents.\n\nKey dates\n- Submission deadline: September 16, 2026 (AoE)\n- Author notification: September 29, 2026 (AoE)\n- Camera-ready: Mid-October 2026\n- Workshop day: December 11, 2026\n\nSubmit via the OpenReview submission site. Part of the AIDrugX workshop lineage at NeurIPS.", "cfp_status": "published", "topics": [ "Agent architectures & reasoning (LLM-based, multimodal, retrieval-augmented agents; long-horizon planners; multi-agent collaborations)", "Harness design & orchestration (biology tool ecosystems, multi-agent routing, long-horizon memory, knowledge grounding)", "Agentic literature & knowledge systems (hypothesis generation, evidence synthesis, claim verification, protocol assistance)", "Generalist vs. specialist model comparisons", "Reinforcement learning for scientific agents", "Recursive self-improvement for bio-specialized models", "Lab-in-the-loop & robotics (closed-loop experimentation, active learning, autonomous laboratories)", "Agents for biological design & discovery (target ID, sequence/molecular design, CRISPR/perturbation design, assay planning)", "Evaluation & benchmarks (faithfulness, reproducibility, calibrated uncertainty, simulator-to-lab transfer)", "Reliability, governance & biosafety" ], "important_dates": [ { "label": "Submission Deadline", "date": "2026-09-16" }, { "label": "Author Notification", "date": "2026-09-29" }, { "label": "Camera-ready", "date": "Mid-October 2026" }, { "label": "Workshop Day", "date": "2026-12-11" } ], "organizers": [], "speakers": [], "host_url": "https://agenticls.github.io", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AgenticLS", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AgenticLS", "location": "Sydney, Australia", "city": "Sydney", "workshop_date": "2026-07-31", "contact": "agenticls@googlegroups.com", "tracks": [ { "key": "AgenticLS", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AgenticLS", "submission_dates_raw": "" } ], "group": "agents", "group_label": "Agents & Agentic Systems", "corpus": "NeurIPS 2026 Agentic AI for Biological Discovery Workshop NeurIPS 2026 Workshop AgenticLS A NeurIPS 2026 workshop on building, benchmarking, and deploying agentic AI systems for life-science discovery — from scientific copilots to closed-loop laboratories — spanning agent design, generalist-vs-specialist model choices, and autonomous lab-in-the-loop experimentation. Agent architectures & reasoning (LLM-based, multimodal, retrieval-augmented agents; long-horizon planners; multi-agent collaborations) Harness design & orchestration (biology tool ecosystems, multi-agent routing, long-horizon memory, knowledge grounding) Agentic literature & knowledge systems (hypothesis generation, evidence synthesis, claim verification, protocol assistance) Generalist vs. specialist model comparisons Reinforcement learning for scientific agents Recursive self-improvement for bio-specialized models Lab-in-the-loop & robotics (closed-loop experimentation, active learning, autonomous laboratories) Agents for biological design & discovery (target ID, sequence/molecular design, CRISPR/perturbation design, assay planning) Evaluation & benchmarks (faithfulness, reproducibility, calibrated uncertainty, simulator-to-lab transfer) Reliability, governance & biosafety Agentic AI for Biological Discovery (AgenticLS)\nFrom scientific copilots to closed-loop labs: building, benchmarking, and deploying agentic AI systems for the next generation of life-science discovery.\n\nAbout the workshop\nThe life sciences are entering an era of agentic AI — systems built on tool-using and reasoning frameworks that go beyond static prediction to read literature, call specialized tools, plan multi-step analyses, propose experiments, and in some cases interact directly with laboratories and robotics. This shift is enabled both by frontier LLMs and by a rapidly growing stack of biology-specialized foundation models for proteins, genomes, and single cells.\n\nYet the field remains strikingly young. There is still little consensus on how to build effective life-science agents, when biology-specialized models are necessary versus when general-purpose LLMs suffice, and how to deploy such agents toward the ultimate goal: accelerating biological discovery and drug development. AgenticLS brings together researchers from machine learning, computational biology, experimental biology, drug discovery, and lab automation to tackle these questions as a building, deployment, and evaluation problem.\n\nCore themes\n1. Building agents for discovery: How should we design agent harnesses, orchestrate multi-agent systems, manage long-horizon memory and context, and integrate biological tool ecosystems? What metrics and rewards guide agents toward novel discoveries?\n2. Generalist vs. specialist: When are frontier general-purpose models (e.g., GPT, Claude) sufficient, and when do biology-specialized models (e.g., AlphaFold, ESM) materially improve planning, reasoning, or downstream outcomes — and can specialists improve via recursive self-improvement?\n3. Autonomous lab-in-the-loop: How should agentic AI autonomously interface with robotics, assay platforms, and human scientists to enable reliable, iterative experimentation in closed-loop laboratories?\n\nScope & topics\nWe welcome submissions across two tracks spanning the full agentic life-science stack.\n\nTrack I — Building Agentic Systems for Life Science\n- Agent architectures & reasoning: LLM-based, multimodal, and retrieval-augmented agents; long-horizon planners; multi-agent scientific collaborations.\n- Harness design & orchestration: tool ecosystems and APIs for biology, multi-agent communication and routing, long-horizon memory and context management, knowledge grounding.\n- Agentic literature & knowledge systems: hypothesis generation, evidence synthesis, claim verification, protocol assistance, ontology-aware and knowledge-graph-enhanced agents.\n- Generalist vs. specialist comparisons: foundation-model backbones, bio-specific instruction tuning, domain adaptation, biological foundation models inside agent pipelines.\n- Reinforcement learning for scientific agents: RL from lab-in-the-loop feedback and reward-function design for discovery.\n- Recursive self-improvement for bio-specialized models: data generation and curation, automated evaluation, feedback-driven refinement, model-in-the-loop workflows.\n\nTrack II — Closed-Loop Discovery & Applications\n- Lab-in-the-loop & robotics: closed-loop experimentation, robotic execution, active learning, adaptive experiment planning, autonomous laboratories that learn from wet-lab feedback.\n- Agents for biological design & discovery: target identification, sequence and molecular design, structure-based design, CRISPR / perturbation design, assay planning, therapeutic optimization.\n- Evaluation & benchmarks: faithfulness, reproducibility, calibrated uncertainty, expert judgment, simulator-to-lab transfer, biological validity, system-level benchmarks.\n- Reliability, governance & biosafety: hallucinated biology, biosecurity-aware safeguards, human oversight, traceability, auditability, and deployment standards for scientific agents.\n\nKey dates\n- Submission deadline: September 16, 2026 (AoE)\n- Author notification: September 29, 2026 (AoE)\n- Camera-ready: Mid-October 2026\n- Workshop day: December 11, 2026\n\nSubmit via the OpenReview submission site. Part of the AIDrugX workshop lineage at NeurIPS." }, { "key": "AgenticOS", "title": "NeurIPS 2026 AgenticOS Workshop", "subtitle": "AgenticOS: Co-designing Systems and ML Foundations of an OS Layer for Agentic AI", "summary": "A workshop bringing the machine learning and systems communities together to define the abstractions, memory hierarchies, scheduling policies, and execution substrates required to make agentic AI reliable, sharable, scalable, and evolvable through a new OS-like layer for agentic AI.", "cfp_full": "Overview\n\nWhy an OS layer for agentic AI?\n\nAgentic AI systems increasingly persist state, plan over long horizons, coordinate multiple models and tools, and even self-evolve. Yet today each capability is rebuilt from scratch inside framework-specific silos. The field needs a new OS-like layer that provides common abstractions for memory, scheduling, routing, governance, and reproducibility.\n\nRecognition of this requirement has led to a proliferation of promising but largely independent efforts, each tackling a different piece of the agentic systems challenge. Designing the needed OS layer requires genuine co-design across the ML and systems communities. The ML community must ask how models should be trained, structured, and exposed as system components, not merely as API endpoints. The systems community must ask what abstractions, memory hierarchies, scheduling policies, and execution substrates are required to make agentic behavior reliable and governable. These questions are fundamentally coupled: the right abstractions depend on model capabilities, and model capabilities depend on system support.\n\n- Scaling beyond pilots: Enterprise agentic systems are scaling beyond pilots but lack robust governance and reproducibility, risking fragmented framework-specific conventions.\n- Protocols at critical mass: Standardization (MCP, A2A) has reached critical mass, enabling definition of system-layer abstractions before defaults solidify.\n- A fragmented field: Relevant research is fragmented across ML, systems, and applications, motivating a dedicated venue to unify vocabulary, benchmarks, and collaboration.\n\nScope - Topics & research questions\n\nThe workshop will explore the following foundational questions:\n- Should foundation models be trained differently to serve as system components?\n- What are the minimal abstractions for building agentic systems?\n- How should agentic memory representations, policies, and cross-layer optimizations be designed?\n- How can agentic workloads be optimized across system layers?\n- How should models be routed and composed under uncertainty?\n- How should long-horizon agentic systems be designed?\n- How can constrained self-evolution preserve safety, reproducibility, and governance?\n- How should agentic systems be evaluated?\n\nOur goal is to establish the conceptual foundations of an operating-system layer for agentic AI by bringing together researchers in machine learning, systems, and scientific applications to identify common abstractions, shared benchmarks, and a long-term research agenda.\n\nCall for submissions\n\nThis workshop focuses on the fundamental research questions at the intersection of ML and systems that must be answered to make agentic AI reliable, sharable, scalable, and evolvable. We invite contributions that address the key questions of the workshop and the following topics, including, but not limited to:\n\nAgentic OS Foundations & Abstractions\n- Minimal, portable abstractions and theoretical principles for building and composing an agentic OS\n- Architectures, training, adaptation, and steering algorithms for foundation models to serve as first-class system components rather than black-box API endpoints\n\nData, Memory & State\n- Agentic memory: representations, retrieval and update policies, consistency, and cross-layer optimization\n- Storage and file systems for agent state, context, and long-lived artifacts\n\nResource Management & Execution\n- Cross-layer optimization for agentic workloads (compute, memory, network, cost, latency)\n- Scheduling and resource management for agentic workloads, including cluster/GPU autoscaling under multi-agent contention\n- Model routing and composition under uncertainty\n- Joint optimization of model composition and context management\n\nLong-Horizon & Self-Evolving Agents\n- Designing long-horizon agentic systems: planning, persistence, and failure recovery\n- Constrained self-evolution: how agents adapt from operational experience while preserving safety and reproducibility\n- Design challenges for agentic systems that support evolution\n\nTrust, Safety & Governance\n- Trust, safety, and security for agentic systems: isolation, access control, and threat models specific to autonomous, tool-using agents\n- Observability and auditability of autonomous agent behavior\n\nEvaluation & Deployment\n- Evaluation methodology, benchmarks, and testbeds for AgenticOS\n- Domain-specific agentic system deployment challenges and solutions\n\nSubmission formats\n\nWe welcome submissions in the following two formats:\n- Extended Abstracts: Extended abstracts may present preliminary work, visionary ideas, or position papers. Length: Up to 2 pages.\n- Regular Papers: Submissions should present original research, real-world experience studies, or analyses of the challenges involved in deploying complex agentic systems. Length: Up to 6 pages.\n\nSubmission guidelines\n\nSubmissions should adhere to the NeurIPS formatting guidelines (download the NeurIPS 2026 template). All submissions must be anonymized for double-blind review. Authors should remove names, affiliations, acknowledgments, and other identifying information from their submissions. References and appendices are not subject to page limits. However, the main paper must be self-contained, and reviewers are not required to consult the appendix.\n\nReview criteria\n\nSubmissions will be evaluated based on technical novelty, interest to the community, lessons learned, and relevance to AgenticOS now or in the future. There are no formal proceedings for this workshop. Accepted submissions will have the option of being published on the workshop website.\n\nImportant dates\n\n- Aug 29, 2026: Paper submission deadline (OpenReview)\n- Sep 29, 2026: Final notifications\n- TBA: Camera-ready paper deadline\n- Dec 12, 2026: Workshop day, NeurIPS 2026, Sydney\n\nAll deadlines are anywhere on Earth (AoE) unless stated otherwise. Dates are subject to change.", "cfp_status": "published", "topics": [ "Agentic OS foundations and abstractions", "Foundation models as first-class system components (architectures, training, adaptation, steering)", "Agentic memory: representations, retrieval/update policies, consistency, cross-layer optimization", "Storage and file systems for agent state, context, and long-lived artifacts", "Cross-layer optimization for agentic workloads (compute, memory, network, cost, latency)", "Scheduling and resource management under multi-agent contention", "Model routing and composition under uncertainty", "Joint optimization of model composition and context management", "Long-horizon agentic systems: planning, persistence, and failure recovery", "Constrained self-evolution preserving safety and reproducibility", "Trust, safety, and security: isolation, access control, threat models", "Observability and auditability of autonomous agent behavior", "Evaluation methodology, benchmarks, and testbeds for AgenticOS", "Domain-specific agentic system deployment challenges and solutions" ], "important_dates": [ { "label": "Paper submission deadline (OpenReview)", "date": "2026-08-29" }, { "label": "Final notifications", "date": "2026-09-29" }, { "label": "Camera-ready paper deadline", "date": "TBA" }, { "label": "Workshop day", "date": "2026-12-12" } ], "organizers": [ "Suparna Bhattacharya (Hewlett Packard Enterprise (HPE Labs))", "Ian Foster (University of Chicago & Argonne National Laboratory)", "Jishen Zhao (University of California, San Diego)", "Cong Wang (Multikernel Technologies)", "Tarun Kumar (Hewlett Packard Enterprise (HPE Labs))" ], "speakers": [ "Ion Stoica (UC Berkeley)", "Hongru Wang (University of Edinburgh)", "Bo Li (University of Illinois Urbana-Champaign)", "Yogesh Simmhan (IISc Bangalore)", "Deshraj Yadav (Mem0)", "Mohamed Wahib (RIKEN R-CCS)", "Manish Gupta (Google DeepMind)" ], "host_url": "https://agentic-fmos.github.io/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AgenticOS", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AgenticOS", "location": "Sydney, Australia", "city": "Sydney", "workshop_date": "2026-12-12", "contact": "tarun.kumar2@hpe.com", "tracks": [ { "key": "AgenticOS", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AgenticOS", "submission_dates_raw": "Submission Deadline: Aug 30 2026 12:30PM UTC-0" } ], "group": "agents", "group_label": "Agents & Agentic Systems", "corpus": "NeurIPS 2026 AgenticOS Workshop AgenticOS: Co-designing Systems and ML Foundations of an OS Layer for Agentic AI A workshop bringing the machine learning and systems communities together to define the abstractions, memory hierarchies, scheduling policies, and execution substrates required to make agentic AI reliable, sharable, scalable, and evolvable through a new OS-like layer for agentic AI. Agentic OS foundations and abstractions Foundation models as first-class system components (architectures, training, adaptation, steering) Agentic memory: representations, retrieval/update policies, consistency, cross-layer optimization Storage and file systems for agent state, context, and long-lived artifacts Cross-layer optimization for agentic workloads (compute, memory, network, cost, latency) Scheduling and resource management under multi-agent contention Model routing and composition under uncertainty Joint optimization of model composition and context management Long-horizon agentic systems: planning, persistence, and failure recovery Constrained self-evolution preserving safety and reproducibility Trust, safety, and security: isolation, access control, threat models Observability and auditability of autonomous agent behavior Evaluation methodology, benchmarks, and testbeds for AgenticOS Domain-specific agentic system deployment challenges and solutions Overview\n\nWhy an OS layer for agentic AI?\n\nAgentic AI systems increasingly persist state, plan over long horizons, coordinate multiple models and tools, and even self-evolve. Yet today each capability is rebuilt from scratch inside framework-specific silos. The field needs a new OS-like layer that provides common abstractions for memory, scheduling, routing, governance, and reproducibility.\n\nRecognition of this requirement has led to a proliferation of promising but largely independent efforts, each tackling a different piece of the agentic systems challenge. Designing the needed OS layer requires genuine co-design across the ML and systems communities. The ML community must ask how models should be trained, structured, and exposed as system components, not merely as API endpoints. The systems community must ask what abstractions, memory hierarchies, scheduling policies, and execution substrates are required to make agentic behavior reliable and governable. These questions are fundamentally coupled: the right abstractions depend on model capabilities, and model capabilities depend on system support.\n\n- Scaling beyond pilots: Enterprise agentic systems are scaling beyond pilots but lack robust governance and reproducibility, risking fragmented framework-specific conventions.\n- Protocols at critical mass: Standardization (MCP, A2A) has reached critical mass, enabling definition of system-layer abstractions before defaults solidify.\n- A fragmented field: Relevant research is fragmented across ML, systems, and applications, motivating a dedicated venue to unify vocabulary, benchmarks, and collaboration.\n\nScope - Topics & research questions\n\nThe workshop will explore the following foundational questions:\n- Should foundation models be trained differently to serve as system components?\n- What are the minimal abstractions for building agentic systems?\n- How should agentic memory representations, policies, and cross-layer optimizations be designed?\n- How can agentic workloads be optimized across system layers?\n- How should models be routed and composed under uncertainty?\n- How should long-horizon agentic systems be designed?\n- How can constrained self-evolution preserve safety, reproducibility, and governance?\n- How should agentic systems be evaluated?\n\nOur goal is to establish the conceptual foundations of an operating-system layer for agentic AI by bringing together researchers in machine learning, systems, and scientific applications to identify common abstractions, shared benchmarks, and a long-term research agenda.\n\nCall for submissions\n\nThis workshop focuses on the fundamental research questions at the intersection of ML and systems that must be answered to make agentic AI reliable, sharable, scalable, and evolvable. We invite contributions that address the key questions of the workshop and the following topics, including, but not limited to:\n\nAgentic OS Foundations & Abstractions\n- Minimal, portable abstractions and theoretical principles for building and composing an agentic OS\n- Architectures, training, adaptation, and steering algorithms for foundation models to serve as first-class system components rather than black-box API endpoints\n\nData, Memory & State\n- Agentic memory: representations, retrieval and update policies, consistency, and cross-layer optimization\n- Storage and file systems for agent state, context, and long-lived artifacts\n\nResource Management & Execution\n- Cross-layer optimization for agentic workloads (compute, memory, network, cost, latency)\n- Scheduling and resource management for agentic workloads, including cluster/GPU autoscaling under multi-agent contention\n- Model routing and composition under uncertainty\n- Joint optimization of model composition and context management\n\nLong-Horizon & Self-Evolving Agents\n- Designing long-horizon agentic systems: planning, persistence, and failure recovery\n- Constrained self-evolution: how agents adapt from operational experience while preserving safety and reproducibility\n- Design challenges for agentic systems that support evolution\n\nTrust, Safety & Governance\n- Trust, safety, and security for agentic systems: isolation, access control, and threat models specific to autonomous, tool-using agents\n- Observability and auditability of autonomous agent behavior\n\nEvaluation & Deployment\n- Evaluation methodology, benchmarks, and testbeds for AgenticOS\n- Domain-specific agentic system deployment challenges and solutions\n\nSubmission formats\n\nWe welcome submissions in the following two formats:\n- Extended Abstracts: Extended abstracts may present preliminary work, visionary ideas, or position papers. Length: Up to 2 pages.\n- Regular Papers: Submissions should present original research, real-world experience studies, or analyses of the challenges involved in deploying complex agentic systems. Length: Up to 6 pages.\n\nSubmission guidelines\n\nSubmissions should adhere to the NeurIPS formatting guidelines (download the NeurIPS 2026 template). All submissions must be anonymized for double-blind review. Authors should remove names, affiliations, acknowledgments, and other identifying information from their submissions. References and appendices are not subject to page limits. However, the main paper must be self-contained, and reviewers are not required to consult the appendix.\n\nReview criteria\n\nSubmissions will be evaluated based on technical novelty, interest to the community, lessons learned, and relevance to AgenticOS now or in the future. There are no formal proceedings for this workshop. Accepted submissions will have the option of being published on the workshop website.\n\nImportant dates\n\n- Aug 29, 2026: Paper submission deadline (OpenReview)\n- Sep 29, 2026: Final notifications\n- TBA: Camera-ready paper deadline\n- Dec 12, 2026: Workshop day, NeurIPS 2026, Sydney\n\nAll deadlines are anywhere on Earth (AoE) unless stated otherwise. Dates are subject to change." }, { "key": "AI4Science", "title": "NeurIPS 2026 AI for Science Workshop: Verification in the Age of AI Scientists", "subtitle": "NeurIPS2026-AI4Science", "summary": "An AI for Science workshop asking how we should trust, judge, and act on AI-generated science when verifiers are imperfect, scarce, or absent—organized around verification in open-ended hypothesis generation, verification under imperfect simulators, and verification under real-world constraints of uncertainty and safety.", "cfp_full": "Verification in the Age of AI Scientists — NeurIPS 2026 AI for Science Workshop\n\nAbout\nAI Scientists now operate at a scale that outpaces human capacity for manual review. Systems such as Sakana's AI Scientist write entire workshop papers end-to-end. Lila Sciences runs autonomous ``AI Science Factories'' that hypothesize, experiment, and iterate without human guidance. FutureHouse's Kosmos and Robin generate thousands of candidate hypotheses in a single run, and Google's Co-Scientist proposes testable experiments at a rate no laboratory can fully evaluate. Each of these systems emphasizes verified results, yet that standard ranges from near-perfect formal proof in mathematics to decade-long clinical trials in medicine, with no shared framework for judging sufficiency across domains. As outputs scale beyond what humans can manually inspect, the question of which results to trust becomes as hard as generating them. The bottleneck for AI for Science is no longer hypothesis generation, it is verification.\n\nOur NeurIPS 2026 workshop, Verification in the Age of AI Scientists, asks how we should trust, judge, and act on AI-generated science when verifiers are imperfect, scarce, or absent. In most sciences the verifier itself is imperfect or prohibitively expensive, and as AI Scientists scale beyond what humans can manually inspect, the central problem becomes which AI outputs deserve our scarce verification budget, and on what evidence we should be willing to act. We organize our discussion around three challenges.\n\nVerification in open-ended hypothesis generation: When AI Scientists propose thousands of candidate hypotheses, only a small fraction can ever be tested, and not all plausible outputs are scientifically meaningful, novel, or worth pursuing. Subtle failure modes such as data leakage, benchmark gaming, and hallucinated citations can make an output look verified without being so, and human taste, intuition, and domain expertise remain essential filters that are not yet well understood as learnable verifiers. This workshop asks how to build verifiers that can separate genuinely novel scientific contributions from convincing artifacts.\n\nVerification under imperfect simulators: Scientific domains differ dramatically in the reliability of their verifiers. In mathematics, formal systems such as Lean provide near-perfect verification, but in biology, force fields fail outside their training distribution and structure prediction has well-documented blind spots, while climate models depend on partial observations and expert judgment. This workshop asks when surrogate verifiers can substitute for ground truth, and on what evidence AI Scientists should be willing to act when the two disagree.\n\nVerification under real-world constraints (uncertainty & safety): In practice, verification is bounded by time, cost, experimental throughput, and safety. A single Phase III clinical trial costs hundreds of millions of dollars and takes a decade, and in extreme weather prediction, rare out-of-distribution events drive evacuation and infrastructure decisions before sufficient evidence can be gathered. This workshop asks how scarce verification resources should be allocated across competing AI-generated hypotheses when downstream decisions affect human lives.\n\nCall for Papers — Scope\nThe workshop seeks submissions on verification of AI-generated science across all scientific disciplines.\n\nSubmission Tracks\n- Track A - Original Research: develop or apply methods for verifying AI-generated science, including learned verifiers, formal methods, surrogate-versus-experiment calibration, uncertainty quantification, active experimental design, and safety-aware deployment.\n- Track B - Position Papers: submissions must present clear, contestable arguments about the epistemics of verification, addressing what constitutes adequate verification, when surrogates can replace ground truth, and gaps in current practices. Evaluation emphasizes argument novelty, engagement with actual scientific domains, and claim falsifiability.\n- Track C - Verifier Systems: papers describing a deployed verifier, including formal, learned, simulator-based, human-AI hybrid, and consensus approaches. Must detail construction, intended use, performance, failure modes, and access methods. Authors must release code and verifier artifacts.\n\nSubmissions\nPlease submit your paper through OpenReview. Our workshop is nonarchival, and accepted papers will be posted on the workshop website. Please use the NeurIPS 2026 LaTeX template; the NeurIPS checklist is not required. Change the template footnote to \"Submitted to/Accepted at/Published in the AI for Science workshop (NeurIPS 2026).\" Submissions should be 4-8 pages, with unlimited references and appendices. Submissions accept original unpublished work, recent journal publications, and works-in-progress. Submit via OpenReview (double-blind review).\n\nKey Dates (Anywhere on Earth)\n- Submission deadline: August 29, 2026 AoE\n- Reviewer period: August 31 – September 1, 2026 AoE\n- Reviewer reviews due: September 17, 2026 AoE\n- Area Chair recommendations due: September 24, 2026 AoE\n- Accept/reject notifications: September 29, 2026 AoE\n- Workshop date: December 11 or 12, 2026", "cfp_status": "published", "topics": [ "Verification in open-ended hypothesis generation", "Verification under imperfect simulators", "Verification under real-world constraints (uncertainty & safety)", "Learned verifiers for AI-generated science", "Formal methods for verification", "Surrogate-versus-experiment calibration", "Uncertainty quantification and active experimental design", "Safety-aware deployment of AI Scientists", "Epistemics of verification (position papers)", "Deployed verifier systems (formal, learned, simulator-based, human-AI hybrid, consensus)" ], "important_dates": [ { "label": "Submission deadline", "date": "2026-08-29" }, { "label": "Reviewer period", "date": "August 31 – September 1, 2026" }, { "label": "Reviewer reviews due", "date": "2026-09-17" }, { "label": "Area Chair recommendations due", "date": "2026-09-24" }, { "label": "Accept/reject notifications", "date": "2026-09-29" }, { "label": "Workshop date", "date": "December 11 or 12, 2026" } ], "organizers": [ "Marinka Zitnik (Harvard)", "Priya Donti (MIT)", "Yuanqi Du (Microsoft Research)", "Ada Fang (Harvard)", "Anvita Bhagavathula (MIT)", "Ana Rivera (MIT)", "Emilien Dupont (Google DeepMind)" ], "speakers": [ "Mario Krenn (University of Tübingen)", "Teresa Head-Gordon (UC Berkeley)", "Anna Scaglione (Cornell)", "Adam Zsolt Wagner (Google DeepMind)", "Charlotte Deane (Oxford)", "Amanda Barnard (Australian National University)", "Marinka Zitnik (Harvard, Panel Moderator)", "David Rolnick (McGill and Mila)", "Rianne van den Berg (Microsoft Research)", "Cheng Soon Ong (CSIRO and Australian National University)", "Lina Yao (UNSW)" ], "host_url": "https://ai4sciencecommunity.github.io/neurips26.html", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AI4Science", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AI4Science", "location": "Sydney, Australia", "city": "Sydney", "workshop_date": "December 11 or 12, 2026", "contact": "ai4scienceneurips2026@googlegroups.com", "tracks": [ { "key": "AI4Science", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AI4Science", "submission_dates_raw": "Submission Deadline: Aug 30 2026 11:59PM UTC-0" } ], "group": "science", "group_label": "AI for Science", "corpus": "NeurIPS 2026 AI for Science Workshop: Verification in the Age of AI Scientists NeurIPS2026-AI4Science An AI for Science workshop asking how we should trust, judge, and act on AI-generated science when verifiers are imperfect, scarce, or absent—organized around verification in open-ended hypothesis generation, verification under imperfect simulators, and verification under real-world constraints of uncertainty and safety. Verification in open-ended hypothesis generation Verification under imperfect simulators Verification under real-world constraints (uncertainty & safety) Learned verifiers for AI-generated science Formal methods for verification Surrogate-versus-experiment calibration Uncertainty quantification and active experimental design Safety-aware deployment of AI Scientists Epistemics of verification (position papers) Deployed verifier systems (formal, learned, simulator-based, human-AI hybrid, consensus) Verification in the Age of AI Scientists — NeurIPS 2026 AI for Science Workshop\n\nAbout\nAI Scientists now operate at a scale that outpaces human capacity for manual review. Systems such as Sakana's AI Scientist write entire workshop papers end-to-end. Lila Sciences runs autonomous ``AI Science Factories'' that hypothesize, experiment, and iterate without human guidance. FutureHouse's Kosmos and Robin generate thousands of candidate hypotheses in a single run, and Google's Co-Scientist proposes testable experiments at a rate no laboratory can fully evaluate. Each of these systems emphasizes verified results, yet that standard ranges from near-perfect formal proof in mathematics to decade-long clinical trials in medicine, with no shared framework for judging sufficiency across domains. As outputs scale beyond what humans can manually inspect, the question of which results to trust becomes as hard as generating them. The bottleneck for AI for Science is no longer hypothesis generation, it is verification.\n\nOur NeurIPS 2026 workshop, Verification in the Age of AI Scientists, asks how we should trust, judge, and act on AI-generated science when verifiers are imperfect, scarce, or absent. In most sciences the verifier itself is imperfect or prohibitively expensive, and as AI Scientists scale beyond what humans can manually inspect, the central problem becomes which AI outputs deserve our scarce verification budget, and on what evidence we should be willing to act. We organize our discussion around three challenges.\n\nVerification in open-ended hypothesis generation: When AI Scientists propose thousands of candidate hypotheses, only a small fraction can ever be tested, and not all plausible outputs are scientifically meaningful, novel, or worth pursuing. Subtle failure modes such as data leakage, benchmark gaming, and hallucinated citations can make an output look verified without being so, and human taste, intuition, and domain expertise remain essential filters that are not yet well understood as learnable verifiers. This workshop asks how to build verifiers that can separate genuinely novel scientific contributions from convincing artifacts.\n\nVerification under imperfect simulators: Scientific domains differ dramatically in the reliability of their verifiers. In mathematics, formal systems such as Lean provide near-perfect verification, but in biology, force fields fail outside their training distribution and structure prediction has well-documented blind spots, while climate models depend on partial observations and expert judgment. This workshop asks when surrogate verifiers can substitute for ground truth, and on what evidence AI Scientists should be willing to act when the two disagree.\n\nVerification under real-world constraints (uncertainty & safety): In practice, verification is bounded by time, cost, experimental throughput, and safety. A single Phase III clinical trial costs hundreds of millions of dollars and takes a decade, and in extreme weather prediction, rare out-of-distribution events drive evacuation and infrastructure decisions before sufficient evidence can be gathered. This workshop asks how scarce verification resources should be allocated across competing AI-generated hypotheses when downstream decisions affect human lives.\n\nCall for Papers — Scope\nThe workshop seeks submissions on verification of AI-generated science across all scientific disciplines.\n\nSubmission Tracks\n- Track A - Original Research: develop or apply methods for verifying AI-generated science, including learned verifiers, formal methods, surrogate-versus-experiment calibration, uncertainty quantification, active experimental design, and safety-aware deployment.\n- Track B - Position Papers: submissions must present clear, contestable arguments about the epistemics of verification, addressing what constitutes adequate verification, when surrogates can replace ground truth, and gaps in current practices. Evaluation emphasizes argument novelty, engagement with actual scientific domains, and claim falsifiability.\n- Track C - Verifier Systems: papers describing a deployed verifier, including formal, learned, simulator-based, human-AI hybrid, and consensus approaches. Must detail construction, intended use, performance, failure modes, and access methods. Authors must release code and verifier artifacts.\n\nSubmissions\nPlease submit your paper through OpenReview. Our workshop is nonarchival, and accepted papers will be posted on the workshop website. Please use the NeurIPS 2026 LaTeX template; the NeurIPS checklist is not required. Change the template footnote to \"Submitted to/Accepted at/Published in the AI for Science workshop (NeurIPS 2026).\" Submissions should be 4-8 pages, with unlimited references and appendices. Submissions accept original unpublished work, recent journal publications, and works-in-progress. Submit via OpenReview (double-blind review).\n\nKey Dates (Anywhere on Earth)\n- Submission deadline: August 29, 2026 AoE\n- Reviewer period: August 31 – September 1, 2026 AoE\n- Reviewer reviews due: September 17, 2026 AoE\n- Area Chair recommendations due: September 24, 2026 AoE\n- Accept/reject notifications: September 29, 2026 AoE\n- Workshop date: December 11 or 12, 2026" }, { "key": "MLxOR", "title": "NeurIPS 2026 Second Workshop on ML×OR: Mathematical Foundations and Operational Integration of Machine Learning for Uncertainty-Aware Decision-Making", "subtitle": "NeurIPS 2026 Workshop MLxOR", "summary": "The second MLxOR workshop, supported by the INFORMS Applied Probability Society, explores the synergy between Machine Learning and Operations Research for data-centric, uncertainty-aware decision-making, with a special emphasis this year on decision-making with GenAI+OR.", "cfp_full": "We are excited to invite submissions to the NeurIPS 2026 Second Workshop on MLxOR: Mathematical Foundations and Operational Integration of Machine Learning for Uncertainty-Aware Decision-Making, organized with support from the INFORMS Applied Probability Society. The workshop will be held in person on December 12 or 13 in Atlanta, GA, USA, as part of the NeurIPS 2026 conference.\n\nLaunched in 2025, the inaugural MLxOR workshop explored the growing synergy between Machine Learning (ML) and Operations Research (OR) and attracted a large number of high-quality submissions and attendees, fostering productive interactions in these interdisciplinary communities. Building on this momentum, MLxOR 2026 will continue to present recent developments, discuss challenges, and publicize emerging research opportunities in data-centric decision-making that leverage both the rapid advancement of ML and the principled methodological rigor of OR. In addition, MLxOR 2026 will emphasize a more focused and timely theme of \"decision-making with GenAI+OR\".\n\nWe welcome submissions that develop new methodologies, provide theoretical insights, or present real-world applications at the intersection of ML and OR. This year's workshop will feature a special emphasis on the integration of Generative AI into decision sciences. Relevant topics include, but are not limited to:\n- Policy design, learning, and evaluation powered by Generative AI and digital twins\n- Decision-focused training of generative models\n- Integration of Generative AI into data-driven optimization and operational decision-making\n- Evaluating catastrophic risks of Generative AI using OR techniques, and conversely addressing rare-event problems using generative modeling\n- Agentic AI for closed-loop decision-making and autonomous operational systems\n\nWe encourage submissions from researchers across ML, OR, applied probability, and statistics, as well as practitioners from industry and public sector organizations. Any works broadly relevant to decision-making via GenAI+OR, in subareas such as reinforcement learning, sequential and adaptive decision-making, stochastic control and optimization, distributional robustness, causal inference, foundation models, simulation and digital twins, and applications in healthcare, logistics, manufacturing, transportation, finance, energy systems, revenue management and other operational domains are all welcome.\n\nIn addition to the presentation of accepted papers, the workshop will feature keynote talks, panel discussions, and plenty of opportunities to interact among participants.\n\nLike last year, we also plan to provide financial support for selected students and junior researchers who contribute to the workshop. More details on the application procedure will be available soon.\n\nCall for Papers\n\nSubmission Guidelines\nSubmission site: https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/MLxOR\nPaper length: Maximum 4 pages for the main body, using the NeurIPS conference format. Unlimited references and supplemental materials are permitted beyond the page limit for the main body. However, reviewers are not obliged to review the supplemental materials.\nNon-anonymity and formatting: Submissions are non-anonymous. Moreover, please use the NeurIPS 2026 paper format: https://neurips.cc/Conferences/2026/CallForPapers (in the provided latex template, please adopt the single-blind format by using the \"sglblindworkshop\" option, and you may drop the NeurIPS Paper Checklist expected for main conference submissions; this checklist is not required for our workshop's submissions).\nEligibility policy: Workshop papers are non-archival. That is, submission to the workshop will not preclude future journal or conference publication. On the other hand, per this year's NeurIPS conference policy, previously published works, including papers accepted to and presented at this year's main NeurIPS conference, are not eligible to appear in a workshop, as the workshop goal is to facilitate dynamic discussion of work in progress and future directions.\n\nAll accepted papers will be presented in one of the poster sessions at the workshop. In addition, several selected papers will be chosen as \"spotlight\" for oral presentations.\n\nMoreover, we will continue and expand the workshop-to-journal pipeline from our inaugural workshop last year, where selected outstanding workshop submissions will be invited to submit a full journal version. This year, we are coordinating with three journals, Stochastic Systems, Mathematics of Operations Research, and Operations Research, for workshop-to-journal conversion. Additional details will be announced as they become available.\n\nImportant Dates\nSubmission deadline: August 31, 2026 (AoE)\nPaper Decision Notification: September 29, 2026 (AoE)\nWorkshop Date and Location: December 12 or 13, 2026, in Atlanta, GA, USA", "cfp_status": "published", "topics": [ "Policy design, learning, and evaluation powered by Generative AI and digital twins", "Decision-focused training of generative models", "Integration of Generative AI into data-driven optimization and operational decision-making", "Evaluating catastrophic risks of Generative AI using OR techniques, and addressing rare-event problems using generative modeling", "Agentic AI for closed-loop decision-making and autonomous operational systems", "Reinforcement learning, sequential and adaptive decision-making", "Stochastic control and optimization", "Distributional robustness", "Causal inference", "Foundation models", "Simulation and digital twins", "Applications in healthcare, logistics, manufacturing, transportation, finance, energy systems, and revenue management" ], "important_dates": [ { "label": "Submission Deadline", "date": "2026-08-31" }, { "label": "Paper Decision Notification", "date": "2026-09-29" }, { "label": "Workshop Date", "date": "December 12 or 13, 2026" } ], "organizers": [ "Minshuo Chen (Northwestern University)", "Jing Dong (Columbia University)", "Henry Lam (Columbia University)", "Karthyek Murthy (University of Southern California)", "Min-hwan Oh (Seoul National University)", "Devavrat Shah (MIT)", "Renyuan Xu (Stanford University)", "Enlu Zhou (Georgia Institute of Technology)" ], "speakers": [ "Erick Delage (HEC Montreal)", "Nathan Kallus (Cornell University)", "David Simchi-Levi (MIT)", "Milind Tambe (Harvard University)", "Masatoshi Uehara (OpenAI)", "Benjamin Van Roy (Stanford University)", "Amy Ward (University of Chicago)", "Yao Xie (Georgia Institute of Technology)", "Linwei Xin (Cornell University)", "Lei Ying (University of Michigan)", "Assaf Zeevi (Columbia University)" ], "host_url": "https://mlxor-2026.github.io/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/MLxOR", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/MLxOR", "location": "Atlanta, USA", "city": "Atlanta", "workshop_date": "2026-12-12", "contact": "neurips.mlxor.workshop@gmail.com", "tracks": [ { "key": "MLxOR", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/MLxOR", "submission_dates_raw": "" } ], "group": "theory", "group_label": "Theory & Methods", "corpus": "NeurIPS 2026 Second Workshop on ML×OR: Mathematical Foundations and Operational Integration of Machine Learning for Uncertainty-Aware Decision-Making NeurIPS 2026 Workshop MLxOR The second MLxOR workshop, supported by the INFORMS Applied Probability Society, explores the synergy between Machine Learning and Operations Research for data-centric, uncertainty-aware decision-making, with a special emphasis this year on decision-making with GenAI+OR. Policy design, learning, and evaluation powered by Generative AI and digital twins Decision-focused training of generative models Integration of Generative AI into data-driven optimization and operational decision-making Evaluating catastrophic risks of Generative AI using OR techniques, and addressing rare-event problems using generative modeling Agentic AI for closed-loop decision-making and autonomous operational systems Reinforcement learning, sequential and adaptive decision-making Stochastic control and optimization Distributional robustness Causal inference Foundation models Simulation and digital twins Applications in healthcare, logistics, manufacturing, transportation, finance, energy systems, and revenue management We are excited to invite submissions to the NeurIPS 2026 Second Workshop on MLxOR: Mathematical Foundations and Operational Integration of Machine Learning for Uncertainty-Aware Decision-Making, organized with support from the INFORMS Applied Probability Society. The workshop will be held in person on December 12 or 13 in Atlanta, GA, USA, as part of the NeurIPS 2026 conference.\n\nLaunched in 2025, the inaugural MLxOR workshop explored the growing synergy between Machine Learning (ML) and Operations Research (OR) and attracted a large number of high-quality submissions and attendees, fostering productive interactions in these interdisciplinary communities. Building on this momentum, MLxOR 2026 will continue to present recent developments, discuss challenges, and publicize emerging research opportunities in data-centric decision-making that leverage both the rapid advancement of ML and the principled methodological rigor of OR. In addition, MLxOR 2026 will emphasize a more focused and timely theme of \"decision-making with GenAI+OR\".\n\nWe welcome submissions that develop new methodologies, provide theoretical insights, or present real-world applications at the intersection of ML and OR. This year's workshop will feature a special emphasis on the integration of Generative AI into decision sciences. Relevant topics include, but are not limited to:\n- Policy design, learning, and evaluation powered by Generative AI and digital twins\n- Decision-focused training of generative models\n- Integration of Generative AI into data-driven optimization and operational decision-making\n- Evaluating catastrophic risks of Generative AI using OR techniques, and conversely addressing rare-event problems using generative modeling\n- Agentic AI for closed-loop decision-making and autonomous operational systems\n\nWe encourage submissions from researchers across ML, OR, applied probability, and statistics, as well as practitioners from industry and public sector organizations. Any works broadly relevant to decision-making via GenAI+OR, in subareas such as reinforcement learning, sequential and adaptive decision-making, stochastic control and optimization, distributional robustness, causal inference, foundation models, simulation and digital twins, and applications in healthcare, logistics, manufacturing, transportation, finance, energy systems, revenue management and other operational domains are all welcome.\n\nIn addition to the presentation of accepted papers, the workshop will feature keynote talks, panel discussions, and plenty of opportunities to interact among participants.\n\nLike last year, we also plan to provide financial support for selected students and junior researchers who contribute to the workshop. More details on the application procedure will be available soon.\n\nCall for Papers\n\nSubmission Guidelines\nSubmission site: https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/MLxOR\nPaper length: Maximum 4 pages for the main body, using the NeurIPS conference format. Unlimited references and supplemental materials are permitted beyond the page limit for the main body. However, reviewers are not obliged to review the supplemental materials.\nNon-anonymity and formatting: Submissions are non-anonymous. Moreover, please use the NeurIPS 2026 paper format: https://neurips.cc/Conferences/2026/CallForPapers (in the provided latex template, please adopt the single-blind format by using the \"sglblindworkshop\" option, and you may drop the NeurIPS Paper Checklist expected for main conference submissions; this checklist is not required for our workshop's submissions).\nEligibility policy: Workshop papers are non-archival. That is, submission to the workshop will not preclude future journal or conference publication. On the other hand, per this year's NeurIPS conference policy, previously published works, including papers accepted to and presented at this year's main NeurIPS conference, are not eligible to appear in a workshop, as the workshop goal is to facilitate dynamic discussion of work in progress and future directions.\n\nAll accepted papers will be presented in one of the poster sessions at the workshop. In addition, several selected papers will be chosen as \"spotlight\" for oral presentations.\n\nMoreover, we will continue and expand the workshop-to-journal pipeline from our inaugural workshop last year, where selected outstanding workshop submissions will be invited to submit a full journal version. This year, we are coordinating with three journals, Stochastic Systems, Mathematics of Operations Research, and Operations Research, for workshop-to-journal conversion. Additional details will be announced as they become available.\n\nImportant Dates\nSubmission deadline: August 31, 2026 (AoE)\nPaper Decision Notification: September 29, 2026 (AoE)\nWorkshop Date and Location: December 12 or 13, 2026, in Atlanta, GA, USA" }, { "key": "TAE", "title": "NeurIPS 2026 Trust-AI-Eval Workshop (TAE): Can We Trust AI Evaluation?", "subtitle": "Robustness, Causality, and Risk in Modern AI Assessment", "summary": "TAE (Trust-AI-Eval) treats AI evaluation itself as an object of study, bringing together work on robustness, causal and measurement validity, auditing, judge reliability, and deployment risk to clarify when AI evaluation results deserve trust.", "cfp_full": "Can We Trust AI Evaluation?\nRobustness, Causality, and Risk in Modern AI Assessment\n\nOverview\n\nCan we trust AI evaluation? Modern AI systems are judged through benchmarks, aggregate scores, and public leaderboards, yet trust in these evaluations is often assumed rather than demonstrated. An evaluation can be precise but measure the wrong construct, stable on a familiar benchmark but brittle on newly collected data, or impressive on a leaderboard while poorly aligned with real-world decisions. Repeated benchmark use, small perturbations, underreported variance, leakage, and contamination can further weaken the evidence behind evaluation claims. The TAE (Trust-AI-Eval): Can We Trust AI Evaluation? workshop treats evaluation itself as an object of study: what is measured, which assumptions connect a protocol to a claim, how uncertainty and failure modes are reported, and when the resulting evidence is strong enough to guide deployment. By bringing together work on robustness, causal and measurement validity, auditing, judge reliability, and deployment risk, the workshop aims to clarify when AI evaluation results deserve trust and how evaluation practices can become more reliable, transparent, and decision-relevant.\n\nWe invite submissions on topics including, but not limited to:\n\n- Uncertainty and robustness: How stable are evaluation conclusions under sampling variation, calibration error, random seeds, data splits, prompts, metrics, evaluator choices, noisy or delayed feedback, tail risk, and worst-case behavior?\n- Benchmark and leaderboard auditing: How do benchmark reuse, contamination, leakage, documentation gaps, lifecycle practice, public incentives, and benchmark-specific optimization affect the trustworthiness of evaluation claims?\n- Black-box auditing: How can AI systems be audited when model internals, training data, or evaluation pipelines are inaccessible, and what behavioral tests, probes, or external evidence can reveal hidden failure modes, contamination, or systematic risk?\n- Measurement and causal validity: What construct is an evaluation protocol intended to measure, what ground truth does it rely on, and what causal, structural, or statistical assumptions connect the protocol to the claim?\n- Stress tests and judge reliability: How should evaluations assess protocol robustness, ambiguous labels, human-, crowd-, and model-judge reliability, and failure modes in evaluation pipelines?\n- Domain coverage and representation: How do imbalances in benchmark suites, such as extensive coverage of coding, mathematics, ethics, and logical reasoning but limited or absent coverage of banking and other regulatory-compliance settings, non-Western cultural contexts, and other underserved domains, affect the validity and generalizability of evaluation claims? How should evaluation portfolios be designed, weighted, and updated to provide representative cross-domain coverage and expose systematic blind spots?\n- Application-domain evaluation: How should evaluation protocols be designed and audited for domain-specific settings such as medicine and healthcare, finance, science, robotics, AI agents, cybersecurity, education, public-sector decision-making, and other high-stakes applications?\n- Deployment risk and governance: When do offline metrics support real-world model selection, safety claims, monitoring, and deployment decisions, and what decision-aware metrics, fairness-accuracy-risk trade-offs, reporting checklists, auditing guidelines, and deployment criteria are needed?\n\nSubmissions will be managed through the OpenReview submission site. Accepted papers will be presented at the in-person poster session.\n\nImportant Dates (Indicative)\n- Paper submission opens: July 30, 2026\n- Paper submission deadline: August 29, 2026 (AoE)\n- Review deadline: September 14, 2026 (AoE)\n- Author notification: September 22, 2026 (AoE)\n- Final program posted: September 27, 2026\n- Workshop: December 11 or 12, 2026", "cfp_status": "published", "topics": [ "Uncertainty and robustness of evaluation conclusions", "Benchmark and leaderboard auditing (reuse, contamination, leakage)", "Black-box auditing of AI systems", "Measurement and causal validity", "Stress tests and judge reliability", "Domain coverage and representation in benchmark suites", "Application-domain evaluation (medicine, finance, science, robotics, agents, cybersecurity, education)", "Deployment risk and governance" ], "important_dates": [ { "label": "Paper submission opens", "date": "2026-07-30" }, { "label": "Paper submission deadline (AoE)", "date": "2026-08-29" }, { "label": "Review deadline (AoE)", "date": "2026-09-14" }, { "label": "Author notification (AoE)", "date": "2026-09-22" }, { "label": "Final program posted", "date": "2026-09-27" }, { "label": "Workshop", "date": "December 11 or 12, 2026" } ], "organizers": [ "Hanxun Huang (University of Melbourne)", "Barbara Tarantino (University of Pavia)", "Paolo Giudici (University of Pavia)", "Xingjun Ma (Fudan University)", "Eduard Hovy (University of Melbourne)", "Sarah M. Erfani (University of Melbourne)" ], "speakers": [ "Bin Yu (University of California, Berkeley)", "Soheil Feizi (University of Maryland)", "Liming Zhu (CSIRO and University of New South Wales)", "James Bailey (Monash University)" ], "host_url": "https://tai-eval.github.io", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/TAE", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/TAE", "location": "Sydney, Australia", "city": "Sydney", "workshop_date": "2026-12-10", "contact": "aiteval2026@gmail.com", "tracks": [ { "key": "TAE", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/TAE", "submission_dates_raw": "" } ], "group": "trust", "group_label": "Trustworthy & Safe AI", "corpus": "NeurIPS 2026 Trust-AI-Eval Workshop (TAE): Can We Trust AI Evaluation? Robustness, Causality, and Risk in Modern AI Assessment TAE (Trust-AI-Eval) treats AI evaluation itself as an object of study, bringing together work on robustness, causal and measurement validity, auditing, judge reliability, and deployment risk to clarify when AI evaluation results deserve trust. Uncertainty and robustness of evaluation conclusions Benchmark and leaderboard auditing (reuse, contamination, leakage) Black-box auditing of AI systems Measurement and causal validity Stress tests and judge reliability Domain coverage and representation in benchmark suites Application-domain evaluation (medicine, finance, science, robotics, agents, cybersecurity, education) Deployment risk and governance Can We Trust AI Evaluation?\nRobustness, Causality, and Risk in Modern AI Assessment\n\nOverview\n\nCan we trust AI evaluation? Modern AI systems are judged through benchmarks, aggregate scores, and public leaderboards, yet trust in these evaluations is often assumed rather than demonstrated. An evaluation can be precise but measure the wrong construct, stable on a familiar benchmark but brittle on newly collected data, or impressive on a leaderboard while poorly aligned with real-world decisions. Repeated benchmark use, small perturbations, underreported variance, leakage, and contamination can further weaken the evidence behind evaluation claims. The TAE (Trust-AI-Eval): Can We Trust AI Evaluation? workshop treats evaluation itself as an object of study: what is measured, which assumptions connect a protocol to a claim, how uncertainty and failure modes are reported, and when the resulting evidence is strong enough to guide deployment. By bringing together work on robustness, causal and measurement validity, auditing, judge reliability, and deployment risk, the workshop aims to clarify when AI evaluation results deserve trust and how evaluation practices can become more reliable, transparent, and decision-relevant.\n\nWe invite submissions on topics including, but not limited to:\n\n- Uncertainty and robustness: How stable are evaluation conclusions under sampling variation, calibration error, random seeds, data splits, prompts, metrics, evaluator choices, noisy or delayed feedback, tail risk, and worst-case behavior?\n- Benchmark and leaderboard auditing: How do benchmark reuse, contamination, leakage, documentation gaps, lifecycle practice, public incentives, and benchmark-specific optimization affect the trustworthiness of evaluation claims?\n- Black-box auditing: How can AI systems be audited when model internals, training data, or evaluation pipelines are inaccessible, and what behavioral tests, probes, or external evidence can reveal hidden failure modes, contamination, or systematic risk?\n- Measurement and causal validity: What construct is an evaluation protocol intended to measure, what ground truth does it rely on, and what causal, structural, or statistical assumptions connect the protocol to the claim?\n- Stress tests and judge reliability: How should evaluations assess protocol robustness, ambiguous labels, human-, crowd-, and model-judge reliability, and failure modes in evaluation pipelines?\n- Domain coverage and representation: How do imbalances in benchmark suites, such as extensive coverage of coding, mathematics, ethics, and logical reasoning but limited or absent coverage of banking and other regulatory-compliance settings, non-Western cultural contexts, and other underserved domains, affect the validity and generalizability of evaluation claims? How should evaluation portfolios be designed, weighted, and updated to provide representative cross-domain coverage and expose systematic blind spots?\n- Application-domain evaluation: How should evaluation protocols be designed and audited for domain-specific settings such as medicine and healthcare, finance, science, robotics, AI agents, cybersecurity, education, public-sector decision-making, and other high-stakes applications?\n- Deployment risk and governance: When do offline metrics support real-world model selection, safety claims, monitoring, and deployment decisions, and what decision-aware metrics, fairness-accuracy-risk trade-offs, reporting checklists, auditing guidelines, and deployment criteria are needed?\n\nSubmissions will be managed through the OpenReview submission site. Accepted papers will be presented at the in-person poster session.\n\nImportant Dates (Indicative)\n- Paper submission opens: July 30, 2026\n- Paper submission deadline: August 29, 2026 (AoE)\n- Review deadline: September 14, 2026 (AoE)\n- Author notification: September 22, 2026 (AoE)\n- Final program posted: September 27, 2026\n- Workshop: December 11 or 12, 2026" }, { "key": "LIGHT", "title": "NeurIPS 2026 Workshop - LIGHT: Deployable Small Foundation Models", "subtitle": "LIGHT", "summary": "LIGHT focuses on the transition from large foundation models to compact, deployable AI systems that remain efficient, governable, and suitable for industrial and regulated environments, bringing together the model-compression and trustworthy-AI communities.", "cfp_full": "LIGHT 2026 - Small, trustworthy and energy-efficient models for real-world deployment.\n\nLIGHT focuses on the transition from large foundation models to compact, deployable AI systems that remain efficient, governable and suitable for industrial and regulated environments.\n\nWorkshop motivation and scope\n\nWhy LIGHT now\n\nFoundation models have transformed artificial intelligence, but their increasing size has made them costly to train, difficult to deploy, and challenging to operate in industrial, edge, and regulated environments. While fine-tuning remains the dominant paradigm for domain adaptation, it does not address the fundamental issues of computational efficiency, energy consumption, governance, and operational trustworthiness. A new research direction is therefore emerging around lightweight foundation models that combine knowledge distillation, compression, quantization, and pruning with advances in trustworthy AI, neuro-symbolic reasoning, and systems engineering. By bringing together these traditionally separate communities, LIGHT aims to advance the next generation of AI systems that are not only accurate, but also efficient, explainable, robust, governable, and ready for real-world deployment.\n\nCompact AI for deployment\nThe workshop addresses the emerging shift from adapting large foundation models to transforming them into deployable AI systems using distillation, compression, quantization and pruning.\n\nTrustworthiness and governance\nSmaller models create new opportunities for explainability, robustness, compliance and alignment with domain-specific requirements, especially in industrial and regulated environments.\n\nCall for Papers\n\nThe LIGHT workshop invites submissions on methods, systems, and applications for lightweight and trustworthy foundation models, welcoming extended abstracts, position papers, system demonstrations, industrial deployment reports, reproducibility contributions and benchmark contributions.\n\nTopics of Interest\n- Distillation of foundation-model capabilities into compact models\n- Small language models and compact multimodal systems\n- Trustworthiness-by-design through model compression\n- Neuro-symbolic extensions for compact foundation models\n- Explainability, reasoning and verification in small models\n- Robustness and safety of compressed AI systems\n- Runtime guarantees for deployable AI\n- Efficient adaptation beyond fine-tuning\n- Edge and resource-constrained deployment\n- Sustainable AI through compact architectures\n- Benchmarks and evaluation methodologies\n- Industrial deployment of compact and trustworthy AI\n\nSubmission\n\nAll submissions will be handled through OpenReview. Each submission will receive at least three reviews from members of the Program Committee. All work must follow the official NeurIPS 2026 author kit using provided LaTeX or Word templates.\n\nPublication Policy\nLIGHT operates as a non-archival workshop, meaning accepted submissions will not appear in archival proceedings. This enables authors to develop work further for future venues. The organizers are pursuing a potential Special Issue with IEEE Transactions on Industrial Informatics, with selected outstanding papers invited for substantially expanded journal submissions under standard peer review.\n\nWorkshop format\n- Research sessions: Three sessions covering distillation, trustworthy compact systems and deployment.\n- Panel discussion: A discussion on whether small models can outperform large models in real-world environments.\n- Poster & networking: An extended interactive session for demos, exchange and collaboration.\n\nImportant Dates\nSubmission deadline: 29 August 2026\nNotification to authors: 29 September 2026\nCamera-ready deadline: October 2026\nWorkshop date: 12-13 December 2026, Paris", "cfp_status": "published", "topics": [ "Distillation of foundation-model capabilities into compact models", "Small language models and compact multimodal systems", "Trustworthiness-by-design through model compression", "Neuro-symbolic extensions for compact foundation models", "Explainability, reasoning and verification in small models", "Robustness and safety of compressed AI systems", "Runtime guarantees for deployable AI", "Efficient adaptation beyond fine-tuning", "Edge and resource-constrained deployment", "Sustainable AI through compact architectures", "Benchmarks and evaluation methodologies", "Industrial deployment of compact and trustworthy AI" ], "important_dates": [ { "label": "Submission Deadline", "date": "2026-08-29" }, { "label": "Notification to Authors", "date": "2026-09-29" }, { "label": "Camera-ready Deadline", "date": "October 2026" }, { "label": "Workshop Date", "date": "December 12-13, 2026" } ], "organizers": [ "Roberta Calegari (University of Bologna)", "Michela Milano (University of Bologna / FBK)", "Dennis Hoppe (High-Performance Computing Center Stuttgart, HLRS)", "Joachim Koehler (Fraunhofer IAIS)" ], "speakers": [], "host_url": "https://almaai-disi-unibo.github.io/neurips2026-light-smallModels/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/LIGHT", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/LIGHT", "location": "Paris, France - colocated with Neurips 2026", "city": "Paris", "workshop_date": "2026-12-11", "contact": "roberta.calegari@unibo.it", "tracks": [ { "key": "LIGHT", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/LIGHT", "submission_dates_raw": "" } ], "group": "fm", "group_label": "Foundation Models & LLMs", "corpus": "NeurIPS 2026 Workshop - LIGHT: Deployable Small Foundation Models LIGHT LIGHT focuses on the transition from large foundation models to compact, deployable AI systems that remain efficient, governable, and suitable for industrial and regulated environments, bringing together the model-compression and trustworthy-AI communities. Distillation of foundation-model capabilities into compact models Small language models and compact multimodal systems Trustworthiness-by-design through model compression Neuro-symbolic extensions for compact foundation models Explainability, reasoning and verification in small models Robustness and safety of compressed AI systems Runtime guarantees for deployable AI Efficient adaptation beyond fine-tuning Edge and resource-constrained deployment Sustainable AI through compact architectures Benchmarks and evaluation methodologies Industrial deployment of compact and trustworthy AI LIGHT 2026 - Small, trustworthy and energy-efficient models for real-world deployment.\n\nLIGHT focuses on the transition from large foundation models to compact, deployable AI systems that remain efficient, governable and suitable for industrial and regulated environments.\n\nWorkshop motivation and scope\n\nWhy LIGHT now\n\nFoundation models have transformed artificial intelligence, but their increasing size has made them costly to train, difficult to deploy, and challenging to operate in industrial, edge, and regulated environments. While fine-tuning remains the dominant paradigm for domain adaptation, it does not address the fundamental issues of computational efficiency, energy consumption, governance, and operational trustworthiness. A new research direction is therefore emerging around lightweight foundation models that combine knowledge distillation, compression, quantization, and pruning with advances in trustworthy AI, neuro-symbolic reasoning, and systems engineering. By bringing together these traditionally separate communities, LIGHT aims to advance the next generation of AI systems that are not only accurate, but also efficient, explainable, robust, governable, and ready for real-world deployment.\n\nCompact AI for deployment\nThe workshop addresses the emerging shift from adapting large foundation models to transforming them into deployable AI systems using distillation, compression, quantization and pruning.\n\nTrustworthiness and governance\nSmaller models create new opportunities for explainability, robustness, compliance and alignment with domain-specific requirements, especially in industrial and regulated environments.\n\nCall for Papers\n\nThe LIGHT workshop invites submissions on methods, systems, and applications for lightweight and trustworthy foundation models, welcoming extended abstracts, position papers, system demonstrations, industrial deployment reports, reproducibility contributions and benchmark contributions.\n\nTopics of Interest\n- Distillation of foundation-model capabilities into compact models\n- Small language models and compact multimodal systems\n- Trustworthiness-by-design through model compression\n- Neuro-symbolic extensions for compact foundation models\n- Explainability, reasoning and verification in small models\n- Robustness and safety of compressed AI systems\n- Runtime guarantees for deployable AI\n- Efficient adaptation beyond fine-tuning\n- Edge and resource-constrained deployment\n- Sustainable AI through compact architectures\n- Benchmarks and evaluation methodologies\n- Industrial deployment of compact and trustworthy AI\n\nSubmission\n\nAll submissions will be handled through OpenReview. Each submission will receive at least three reviews from members of the Program Committee. All work must follow the official NeurIPS 2026 author kit using provided LaTeX or Word templates.\n\nPublication Policy\nLIGHT operates as a non-archival workshop, meaning accepted submissions will not appear in archival proceedings. This enables authors to develop work further for future venues. The organizers are pursuing a potential Special Issue with IEEE Transactions on Industrial Informatics, with selected outstanding papers invited for substantially expanded journal submissions under standard peer review.\n\nWorkshop format\n- Research sessions: Three sessions covering distillation, trustworthy compact systems and deployment.\n- Panel discussion: A discussion on whether small models can outperform large models in real-world environments.\n- Poster & networking: An extended interactive session for demos, exchange and collaboration.\n\nImportant Dates\nSubmission deadline: 29 August 2026\nNotification to authors: 29 September 2026\nCamera-ready deadline: October 2026\nWorkshop date: 12-13 December 2026, Paris" }, { "key": "ML4Molecules", "title": "NeurIPS 2026 Workshop on Agentic Systems for Molecular Sciences", "subtitle": "ML4Molecules 2026", "summary": "A NeurIPS 2026 workshop on Agentic Systems for Molecular Sciences that examines the contrast between agentic ambition and methodological fragility, inviting work across the full stack from foundational ML methods to end-to-end scientific agents, with explicit space for negative results, rigorous baselines, and benchmarks.", "cfp_full": "Agentic Systems for Molecular Sciences\nFrom in-silico chemists and closed-loop labs to failures at counting carbons.\nA NeurIPS 2026 Workshop · Paris · Dec 12–13\n\nAgentic large language model systems are moving from conversational and coding assistants to co-scientists in the molecular sciences, such as planning syntheses, calling domain tools, and closing experimental loops with laboratory automation. Recent demonstrations span autonomous drug repurposing, retrosynthesis, and genome-wide virtual screening, all built on stacks of learned representations, predictors, and simulators. Yet careful benchmarking has exposed striking failures at seemingly simple tasks: tool-augmented agents reach only around 50% accuracy on chemical cost estimation, chemistry language models fail systematic symbolic reasoning on molecular graphs, and single-cell foundation models for perturbation prediction do not outperform linear baselines. This workshop takes the contrast between agentic ambition and methodological fragility as its starting point with explicit space for negative results, rigorous baselines, and benchmark contributions alongside methodological advances.\n\nCall for Papers\nWe invite submissions to Agentic Systems for Molecular Sciences, a NeurIPS 2026 workshop taking place in Paris on 12 or 13 December.\n\nAgentic systems are only as good as the representations, predictors, generative models, and simulators they orchestrate. Progress on the agentic frontier depends on progress in the underlying machine learning methods, and on honest evaluation of both. We invite submissions across the full stack, from foundational methods to end-to-end agents, including negative results, careful baselines, and benchmark contributions alongside methodological advances. We welcome participation from machine learning researchers, chemists, biologists, materials scientists, and interdisciplinary practitioners, and encourage submissions from early-career and underrepresented authors.\n\nTopics of interest\nTopics include, but are not limited to:\n- Agentic and multi-agent systems for the molecular sciences; tool use, planning, and orchestration\n- Benchmarks, reproducibility, negative results, and rigorous evaluation of scientific agents and their components\n- Grounding LLMs via chemical databases, simulators, and experimental feedback\n- Closed-loop / self-driving labs, active learning, and Bayesian optimization for experimental design\n- Foundation models for chemistry, biology, and materials, and their benchmarking\n- Representation learning and generative models for molecules, proteins, reactions, and materials\n- Bioactivity, ADME, toxicity, and molecular property prediction\n- Reaction prediction, retrosynthesis, and synthesis planning\n- Structure-based and ligand-based virtual screening; docking, scoring, and protein–ligand / protein–protein interaction prediction\n- Perturbation modelling, batch effect correction, domain adaptation and single-cell prediction\n- Machine-learned interatomic potentials, force fields, molecular dynamics, and geometric / equivariant / physics-informed deep learning\n\nContributions that clarify what current methods can and cannot do, including negative results, careful ablations, and rigorous baselines, are especially welcome.\n\nSubmission guidelines\nFormat. Submissions must use the NeurIPS 2026 workshop LaTeX template. The main text is limited to 5 content pages, including all figures and tables. References and appendices do not count toward the page limit, but reviewers are not obliged to read supplementary material — the main text must be self-contained. Maximum file size: 50 MB.\nAnonymity. Reviewing is double-blind. Submissions must be anonymized, with author names, affiliations, and acknowledgments removed, and prior work cited in the third person.\nSubmission site. Submissions are managed via OpenReview. Papers remain private during review. All authors should maintain up-to-date OpenReview profiles for conflict-of-interest management and paper matching.\nNon-archival policy. The workshop is non-archival. Papers under review elsewhere are welcome, and accepted papers may be published at other venues afterward. Work already published at NeurIPS or other archival ML venues should not be submitted; substantial extensions of prior non-ML-venue work are eligible.\nPresentation. Accepted contributions are presented as posters, with a subset selected for contributed talks. Contributed-talk slots are reserved for early-career first authors. A Best Paper Award will be given for the strongest overall contribution.\n\nImportant Dates\n- Call for papers: mid-July 2026\n- Submission deadline: 29 August 2026\n- Reviewing: 30 August – 25 September 2026\n- Author notification: 29 September 2026\n- Workshop: 12–13 December 2026 · Paris", "cfp_status": "published", "topics": [ "Agentic and multi-agent systems for the molecular sciences; tool use, planning, and orchestration", "Benchmarks, reproducibility, negative results, and rigorous evaluation of scientific agents", "Grounding LLMs via chemical databases, simulators, and experimental feedback", "Closed-loop / self-driving labs, active learning, and Bayesian optimization for experimental design", "Foundation models for chemistry, biology, and materials, and their benchmarking", "Representation learning and generative models for molecules, proteins, reactions, and materials", "Bioactivity, ADME, toxicity, and molecular property prediction", "Reaction prediction, retrosynthesis, and synthesis planning", "Structure-based and ligand-based virtual screening; docking, scoring, and interaction prediction", "Perturbation modelling, batch effect correction, domain adaptation and single-cell prediction", "Machine-learned interatomic potentials, force fields, molecular dynamics, and geometric/equivariant/physics-informed deep learning" ], "important_dates": [ { "label": "Call for papers", "date": "2026-07" }, { "label": "Submission deadline", "date": "2026-08-29" }, { "label": "Reviewing", "date": "2026-08-30 to 2026-09-25" }, { "label": "Author notification", "date": "2026-09-29" }, { "label": "Workshop", "date": "2026-12-12 to 2026-12-13" } ], "organizers": [ "Nadine Schneider (Novartis)", "Günter Klambauer (ELLIS Unit Linz & Johannes Kepler University Linz)", "Ola Engkvist (AstraZeneca & Chalmers University of Technology)", "Marwin Segler (Microsoft Research)", "Sohvi Luukkonen (ELLIS Unit Linz & Johannes Kepler University Linz)" ], "speakers": [ "Andrew White (FutureHouse)", "Francesca Grisoni (TU Eindhoven)", "Yanyan Lan (Tsinghua University)", "Gábor Csányi (University of Cambridge & Max Planck Institute for Polymer Research)", "Jean-Philippe Vert (Bioptimus)", "Philippe Schwaller (EPFL)", "Janine George (FSU Jena & BAM Berlin)" ], "host_url": "https://moleculediscovery.github.io/workshop2026/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/ML4Molecules", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/ML4Molecules", "location": "Paris, France", "city": "Paris", "workshop_date": "2026-12-12", "contact": "ml4molecules@ml.jku.at", "tracks": [ { "key": "ML4Molecules", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/ML4Molecules", "submission_dates_raw": "" } ], "group": "agents", "group_label": "Agents & Agentic Systems", "corpus": "NeurIPS 2026 Workshop on Agentic Systems for Molecular Sciences ML4Molecules 2026 A NeurIPS 2026 workshop on Agentic Systems for Molecular Sciences that examines the contrast between agentic ambition and methodological fragility, inviting work across the full stack from foundational ML methods to end-to-end scientific agents, with explicit space for negative results, rigorous baselines, and benchmarks. Agentic and multi-agent systems for the molecular sciences; tool use, planning, and orchestration Benchmarks, reproducibility, negative results, and rigorous evaluation of scientific agents Grounding LLMs via chemical databases, simulators, and experimental feedback Closed-loop / self-driving labs, active learning, and Bayesian optimization for experimental design Foundation models for chemistry, biology, and materials, and their benchmarking Representation learning and generative models for molecules, proteins, reactions, and materials Bioactivity, ADME, toxicity, and molecular property prediction Reaction prediction, retrosynthesis, and synthesis planning Structure-based and ligand-based virtual screening; docking, scoring, and interaction prediction Perturbation modelling, batch effect correction, domain adaptation and single-cell prediction Machine-learned interatomic potentials, force fields, molecular dynamics, and geometric/equivariant/physics-informed deep learning Agentic Systems for Molecular Sciences\nFrom in-silico chemists and closed-loop labs to failures at counting carbons.\nA NeurIPS 2026 Workshop · Paris · Dec 12–13\n\nAgentic large language model systems are moving from conversational and coding assistants to co-scientists in the molecular sciences, such as planning syntheses, calling domain tools, and closing experimental loops with laboratory automation. Recent demonstrations span autonomous drug repurposing, retrosynthesis, and genome-wide virtual screening, all built on stacks of learned representations, predictors, and simulators. Yet careful benchmarking has exposed striking failures at seemingly simple tasks: tool-augmented agents reach only around 50% accuracy on chemical cost estimation, chemistry language models fail systematic symbolic reasoning on molecular graphs, and single-cell foundation models for perturbation prediction do not outperform linear baselines. This workshop takes the contrast between agentic ambition and methodological fragility as its starting point with explicit space for negative results, rigorous baselines, and benchmark contributions alongside methodological advances.\n\nCall for Papers\nWe invite submissions to Agentic Systems for Molecular Sciences, a NeurIPS 2026 workshop taking place in Paris on 12 or 13 December.\n\nAgentic systems are only as good as the representations, predictors, generative models, and simulators they orchestrate. Progress on the agentic frontier depends on progress in the underlying machine learning methods, and on honest evaluation of both. We invite submissions across the full stack, from foundational methods to end-to-end agents, including negative results, careful baselines, and benchmark contributions alongside methodological advances. We welcome participation from machine learning researchers, chemists, biologists, materials scientists, and interdisciplinary practitioners, and encourage submissions from early-career and underrepresented authors.\n\nTopics of interest\nTopics include, but are not limited to:\n- Agentic and multi-agent systems for the molecular sciences; tool use, planning, and orchestration\n- Benchmarks, reproducibility, negative results, and rigorous evaluation of scientific agents and their components\n- Grounding LLMs via chemical databases, simulators, and experimental feedback\n- Closed-loop / self-driving labs, active learning, and Bayesian optimization for experimental design\n- Foundation models for chemistry, biology, and materials, and their benchmarking\n- Representation learning and generative models for molecules, proteins, reactions, and materials\n- Bioactivity, ADME, toxicity, and molecular property prediction\n- Reaction prediction, retrosynthesis, and synthesis planning\n- Structure-based and ligand-based virtual screening; docking, scoring, and protein–ligand / protein–protein interaction prediction\n- Perturbation modelling, batch effect correction, domain adaptation and single-cell prediction\n- Machine-learned interatomic potentials, force fields, molecular dynamics, and geometric / equivariant / physics-informed deep learning\n\nContributions that clarify what current methods can and cannot do, including negative results, careful ablations, and rigorous baselines, are especially welcome.\n\nSubmission guidelines\nFormat. Submissions must use the NeurIPS 2026 workshop LaTeX template. The main text is limited to 5 content pages, including all figures and tables. References and appendices do not count toward the page limit, but reviewers are not obliged to read supplementary material — the main text must be self-contained. Maximum file size: 50 MB.\nAnonymity. Reviewing is double-blind. Submissions must be anonymized, with author names, affiliations, and acknowledgments removed, and prior work cited in the third person.\nSubmission site. Submissions are managed via OpenReview. Papers remain private during review. All authors should maintain up-to-date OpenReview profiles for conflict-of-interest management and paper matching.\nNon-archival policy. The workshop is non-archival. Papers under review elsewhere are welcome, and accepted papers may be published at other venues afterward. Work already published at NeurIPS or other archival ML venues should not be submitted; substantial extensions of prior non-ML-venue work are eligible.\nPresentation. Accepted contributions are presented as posters, with a subset selected for contributed talks. Contributed-talk slots are reserved for early-career first authors. A Best Paper Award will be given for the strongest overall contribution.\n\nImportant Dates\n- Call for papers: mid-July 2026\n- Submission deadline: 29 August 2026\n- Reviewing: 30 August – 25 September 2026\n- Author notification: 29 September 2026\n- Workshop: 12–13 December 2026 · Paris" }, { "key": "AI4DD", "title": "NeurIPS 2026 Workshop on AI for Drug Discovery: Bridging the Translation Gap", "subtitle": "NeurIPS2026-AI4DD", "summary": "A NeurIPS 2026 workshop addressing the translation gap between computational success and real-world impact in AI-driven drug discovery, connecting ML researchers with computational chemists, structural biologists, experimental scientists, and industry practitioners to make AI reliable, interpretable, experimentally actionable, and translatable.", "cfp_full": "About the Workshop\nArtificial intelligence is rapidly reshaping drug discovery, with advances in deep learning, foundation models, generative modeling, structure prediction, and scientific agents showing strong potential across molecular property prediction, ligand and protein design, biomolecular interaction modeling, and experimental prioritization. Yet a substantial translation gap remains between computational success and real-world impact: strong performance on static, curated benchmarks does not necessarily persist under prospective experiments, new targets, new chemical series, assay shifts, or operational constraints.\n\nThis workshop brings the NeurIPS community together around this challenge — connecting machine learning researchers with computational chemists, structural biologists, experimental scientists, and industry practitioners to define what it means for AI in drug discovery to be not only accurate on benchmarks, but reliable, interpretable, experimentally actionable, and ultimately translatable to real therapeutic development.\n\nCall for Papers\nWe invite submissions of original research, recently published work in scientific journals, and work in progress on AI for drug discovery, including but not limited to the following topics.\n\nScope & Topics\n- From Benchmarks to Closed-Loop Translation — realistic benchmark design, external and prospective validation, retrospective–prospective discrepancies, negative and inconclusive results, failure analysis, evidence standards, reproducibility, and integration into design–make–test–learn workflows.\n- Learning under Scarce, Biased, and Shifting Data — few-shot, zero-shot, transfer, active, and physics-informed learning; noisy or censored assays; missing modalities; cross-laboratory and cross-target distribution shift; adaptation from public datasets to real discovery campaigns.\n- Scientific Reasoning and AI Co-Scientists — grounded hypothesis generation, experiment planning, multimodal reasoning, tool use, provenance, verification, failure recovery, and prospective evaluation of long-horizon scientific workflows.\n- Trustworthy and Decision-Aware AI — uncertainty quantification, calibration, conformal and selective prediction, out-of-distribution detection, causal and mechanistic interpretation, abstention, decision-aware metrics.\n- Multi-Objective and Resource-Constrained Discovery — cost-aware optimization and experimental value of information under conflicting objectives such as potency, selectivity, toxicity, synthesizability, pharmacokinetics, developability, and limited wet-lab budgets.\n\nSubmission Instructions\nWe welcome full-paper (up to 5 pages) and short-paper (up to 2 pages) submissions, excluding references and appendix. Review is double-blind and conducted through OpenReview. Please use the NeurIPS 2026 LaTeX template; the NeurIPS checklist is not required. Please change the template footnote to \"Submitted to in the AI for Drug Discovery workshop (NeurIPS 2026).\" This workshop is non-archival and does not publish proceedings. Contributed talks and Best Paper Awards will be selected based on review scores and recommendations from the reviewers, followed by discussion among the workshop chairs.\n\nImportant Dates\nAll deadlines are 11:59 PM AoE (Anywhere on Earth), tentative pending final NeurIPS scheduling.\n- Call for papers: 15 July 2026\n- Submission deadline: 29 August 2026\n- Reviews due: 15 September 2026\n- Author discussion & meta-review: 16–21 September 2026\n- Acceptance notification: 23 September 2026\n- Camera-ready materials: 15 October 2026", "cfp_status": "published", "topics": [ "From Benchmarks to Closed-Loop Translation: realistic benchmark design, external and prospective validation, failure analysis, reproducibility, design–make–test–learn workflows", "Learning under Scarce, Biased, and Shifting Data: few-shot, zero-shot, transfer, active, and physics-informed learning; distribution shift", "Scientific Reasoning and AI Co-Scientists: hypothesis generation, experiment planning, multimodal reasoning, tool use, verification", "Trustworthy and Decision-Aware AI: uncertainty quantification, calibration, conformal and selective prediction, OOD detection, causal interpretation", "Multi-Objective and Resource-Constrained Discovery: cost-aware optimization under potency, selectivity, toxicity, synthesizability, pharmacokinetics constraints" ], "important_dates": [ { "label": "Call for Papers", "date": "2026-07-15" }, { "label": "Submission Deadline", "date": "2026-08-29" }, { "label": "Reviews Due", "date": "2026-09-15" }, { "label": "Author Discussion & Meta-Review", "date": "2026-09-16 to 2026-09-21" }, { "label": "Acceptance Notification", "date": "2026-09-23" }, { "label": "Camera-Ready Materials", "date": "2026-10-15" } ], "organizers": [ "Ivor Tsang (CFAR, IAIC, A*STAR · NTU · UTS)", "Yinghua Yao (CFAR, IAIC, A*STAR)", "Yu Xie (Microsoft Research AI for Science, Berlin)", "Steve Ling (University of Technology Sydney)", "Matthew Jacobson (AIDD, A*STAR) — Advisory Board", "Chuan-Sheng Foo (Calico Life Sciences) — Advisory Board", "Yew Soon Ong (Nanyang Technological University) — Advisory Board", "Andrea Zerio (CFAR, IAIC, A*STAR · Aalborg University) — Publicity Chair" ], "speakers": [ "Le Song (GenBio AI & MBZUAI)", "Matt Onsum (Calico Life Sciences)", "Yanyan Lan (Tsinghua University)", "Xixian Chen (SIFBI, A*STAR)", "Cheng He (Mirxes)", "Matthew Jacobson (AIDD, A*STAR)" ], "host_url": "https://ai4dd-neurips2026.github.io/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AI4DD", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AI4DD", "location": "Sydney, Australia", "city": "Sydney", "workshop_date": "2026-12-10", "contact": "ai4dd-organizer@googlegroups.com", "tracks": [ { "key": "AI4DD", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AI4DD", "submission_dates_raw": "" } ], "group": "science", "group_label": "AI for Science", "corpus": "NeurIPS 2026 Workshop on AI for Drug Discovery: Bridging the Translation Gap NeurIPS2026-AI4DD A NeurIPS 2026 workshop addressing the translation gap between computational success and real-world impact in AI-driven drug discovery, connecting ML researchers with computational chemists, structural biologists, experimental scientists, and industry practitioners to make AI reliable, interpretable, experimentally actionable, and translatable. From Benchmarks to Closed-Loop Translation: realistic benchmark design, external and prospective validation, failure analysis, reproducibility, design–make–test–learn workflows Learning under Scarce, Biased, and Shifting Data: few-shot, zero-shot, transfer, active, and physics-informed learning; distribution shift Scientific Reasoning and AI Co-Scientists: hypothesis generation, experiment planning, multimodal reasoning, tool use, verification Trustworthy and Decision-Aware AI: uncertainty quantification, calibration, conformal and selective prediction, OOD detection, causal interpretation Multi-Objective and Resource-Constrained Discovery: cost-aware optimization under potency, selectivity, toxicity, synthesizability, pharmacokinetics constraints About the Workshop\nArtificial intelligence is rapidly reshaping drug discovery, with advances in deep learning, foundation models, generative modeling, structure prediction, and scientific agents showing strong potential across molecular property prediction, ligand and protein design, biomolecular interaction modeling, and experimental prioritization. Yet a substantial translation gap remains between computational success and real-world impact: strong performance on static, curated benchmarks does not necessarily persist under prospective experiments, new targets, new chemical series, assay shifts, or operational constraints.\n\nThis workshop brings the NeurIPS community together around this challenge — connecting machine learning researchers with computational chemists, structural biologists, experimental scientists, and industry practitioners to define what it means for AI in drug discovery to be not only accurate on benchmarks, but reliable, interpretable, experimentally actionable, and ultimately translatable to real therapeutic development.\n\nCall for Papers\nWe invite submissions of original research, recently published work in scientific journals, and work in progress on AI for drug discovery, including but not limited to the following topics.\n\nScope & Topics\n- From Benchmarks to Closed-Loop Translation — realistic benchmark design, external and prospective validation, retrospective–prospective discrepancies, negative and inconclusive results, failure analysis, evidence standards, reproducibility, and integration into design–make–test–learn workflows.\n- Learning under Scarce, Biased, and Shifting Data — few-shot, zero-shot, transfer, active, and physics-informed learning; noisy or censored assays; missing modalities; cross-laboratory and cross-target distribution shift; adaptation from public datasets to real discovery campaigns.\n- Scientific Reasoning and AI Co-Scientists — grounded hypothesis generation, experiment planning, multimodal reasoning, tool use, provenance, verification, failure recovery, and prospective evaluation of long-horizon scientific workflows.\n- Trustworthy and Decision-Aware AI — uncertainty quantification, calibration, conformal and selective prediction, out-of-distribution detection, causal and mechanistic interpretation, abstention, decision-aware metrics.\n- Multi-Objective and Resource-Constrained Discovery — cost-aware optimization and experimental value of information under conflicting objectives such as potency, selectivity, toxicity, synthesizability, pharmacokinetics, developability, and limited wet-lab budgets.\n\nSubmission Instructions\nWe welcome full-paper (up to 5 pages) and short-paper (up to 2 pages) submissions, excluding references and appendix. Review is double-blind and conducted through OpenReview. Please use the NeurIPS 2026 LaTeX template; the NeurIPS checklist is not required. Please change the template footnote to \"Submitted to in the AI for Drug Discovery workshop (NeurIPS 2026).\" This workshop is non-archival and does not publish proceedings. Contributed talks and Best Paper Awards will be selected based on review scores and recommendations from the reviewers, followed by discussion among the workshop chairs.\n\nImportant Dates\nAll deadlines are 11:59 PM AoE (Anywhere on Earth), tentative pending final NeurIPS scheduling.\n- Call for papers: 15 July 2026\n- Submission deadline: 29 August 2026\n- Reviews due: 15 September 2026\n- Author discussion & meta-review: 16–21 September 2026\n- Acceptance notification: 23 September 2026\n- Camera-ready materials: 15 October 2026" }, { "key": "AI4MetaScience", "title": "NeurIPS 2026 Workshop on AI for Meta‑Science: Scaling and Organizing Science in the Age of AI", "subtitle": "AI4MetaScience — Scaling and Organizing Science in the Age of AI Scientists", "summary": "A NeurIPS 2026 workshop treating the scientific enterprise itself as an object of study and intervention, establishing AI for meta-science as a distinct research area at the intersection of machine learning, science policy, and the study of science — focused on scaling evaluation, verification, and governance alongside AI-accelerated scientific production.", "cfp_full": "About the Workshop\nAI is rapidly transforming how scientific work is produced. Beyond accelerating computational workflows, AI systems increasingly support or automate major parts of research: hypothesis generation, experiment planning, analysis, writing, and even elements of wet-lab automation. As AI scientists and AI-augmented pipelines reduce the marginal cost of producing scientific artifacts, the scientific ecosystem faces a new bottleneck: evaluation, verification, and governance.\n\nWhile the production side of science has scaled dramatically, quality control (peer review, reproducibility norms, publication criteria, incentive structures, and funding allocation) is lagging behind. Record submission numbers at major AI conferences and a growing reviewer crisis make this gap visible year after year. This workshop treats the scientific enterprise itself as an object of study and intervention, and aims to establish AI for meta-science as a distinct research area at the intersection of machine learning, science policy, and the study of science. Questions we will tackle include:\n- Scaling evaluation alongside production: how can peer review, reproducibility checking, and research evaluation keep up when AI scientists can produce papers at scale, and what roles should AI reviewers and verification tools play?\n- Quality control and integrity: how do we detect and handle AI-generated content, manipulation, and \"AI slop\" in scholarly communication, while responsibly incorporating AI assistance into authoring and reviewing?\n- Reorganizing the scientific enterprise: what counts as a publication now? How should credit, careers, incentives, funding, and governance adapt to AI-accelerated science, in academia and industry alike?\n\nWho should attend\nWe bring together researchers from meta-science, philosophy of science, AI, peer review, and evaluation science with conference organizers, journal editors, and arXiv moderators. If you build AI systems for the scientific process, study the process itself, or run parts of it, this workshop is for you.\n\nFormat and outcomes\nThe program combines invited talks, contributed talks, a poster session, a panel discussion, and unconference-style breakout sessions where attendees propose their own topics for collaborative research sprints and community building. The workshop aims to produce a position paper with a set of recommendations based on participant input, to establish guidelines for ML communities on dealing with the peer-review crisis and AI in the scientific process, and to launch a lasting working group on AI for meta-science.\n\nCall for Papers\nThe workshop addresses the use, evaluation, and governance of AI systems that shape the scientific process itself: peer review, reproducibility, research evaluation, publication incentives, and funding and governance mechanisms.\n\nTwo Submission Tracks:\n- Technical Track: Methods, systems, benchmarks, datasets, and empirical analyses covering AI-assisted peer review, autonomous research agents, reproducibility verification, AI-generated content detection, and reviewer assignment systems.\n- Position Track: Argumentative papers on how the scientific ecosystem should evolve. These need not be technical and welcome perspectives from philosophy, science studies, law, and policy. Titles must begin with \"Position:\".\n\nSubmission Requirements\nSubmissions are made through the workshop's OpenReview page. Papers should use the NeurIPS 2026 LaTeX template, with the footnote changed to \"Submitted to AI for Meta-Science workshop (NeurIPS 2026)\". The NeurIPS paper checklist is not needed. Submissions are either 4 pages (concise technical contributions) or 8 pages (more substantial contributions, for example position papers), excluding references, with unlimited appendix. We welcome papers in two tracks: a technical track and a position paper track; previously published relevant work is also welcome for presentation. Novel work requires double-blind anonymization; previously published work does not. Authors may briefly disclose AI's role in producing the submission.\n\nReview Process\nSubmissions receive minimum two reviews (three preferred). Reviewers must read full main text and may use AI assistance. Decisions fall into three categories: reject, accept (poster), or accept (oral). The workshop is explicitly non-archival, allowing arXiv posting before or after. Dual submission with other venues is permitted, and previously published work is welcome.\n\nImportant Dates (Anywhere on Earth)\n- Submissions open: July 27, 2026\n- Paper submission deadline: August 29, 2026\n- Accept/reject notification: September 29, 2026\n- Camera-ready deadline: November 2026 (exact date TBA)\n- Workshop: December 2026 (exact day TBA) at NeurIPS 2026, Paris", "cfp_status": "published", "topics": [ "Scaling evaluation alongside production: AI-assisted peer review, reproducibility checking, research evaluation", "Quality control and integrity: detecting AI-generated content, manipulation, and 'AI slop'; responsible AI assistance in authoring and reviewing", "Reorganizing the scientific enterprise: publication criteria, credit, careers, incentives, funding, and governance under AI-accelerated science", "Autonomous research agents and AI scientists", "Reproducibility verification and AI-generated content detection", "Reviewer assignment systems", "Position papers on the future of the scientific ecosystem (philosophy, science studies, law, policy)" ], "important_dates": [ { "label": "Submissions Open", "date": "2026-07-27" }, { "label": "Paper Submission Deadline", "date": "2026-08-29" }, { "label": "Accept/Reject Notification", "date": "2026-09-29" }, { "label": "Camera-Ready Deadline", "date": "2026-11" }, { "label": "Workshop", "date": "2026-12-11" } ], "organizers": [ "Prabhant Singh (Microsoft, Amsterdam)", "Khuong T G Hieu (Université Paris-Saclay)", "Hilde Weerts (Eindhoven University of Technology)", "Lele Cao (Microsoft Gaming, King AI Labs)", "Joaquin Vanschoren (Eindhoven University of Technology)" ], "speakers": [ "Isabelle Guyon (Google DeepMind, Paris)", "Vlasta Sikimić (Eindhoven University of Technology)", "Joydeep Biswas (The University of Texas at Austin)", "Jack Stilgoe (University College London)", "Autumn Toney-Wails (UNU-MERIT, Maastricht University) — panel moderator" ], "host_url": "https://ai4metascience.org/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AI4MetaScience", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AI4MetaScience", "location": "Paris, France", "city": "Paris", "workshop_date": "2026-12-11", "contact": "ai4metascience@googlegroups.com", "tracks": [ { "key": "AI4MetaScience", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AI4MetaScience", "submission_dates_raw": "" } ], "group": "society", "group_label": "Society, Applications & Community", "corpus": "NeurIPS 2026 Workshop on AI for Meta‑Science: Scaling and Organizing Science in the Age of AI AI4MetaScience — Scaling and Organizing Science in the Age of AI Scientists A NeurIPS 2026 workshop treating the scientific enterprise itself as an object of study and intervention, establishing AI for meta-science as a distinct research area at the intersection of machine learning, science policy, and the study of science — focused on scaling evaluation, verification, and governance alongside AI-accelerated scientific production. Scaling evaluation alongside production: AI-assisted peer review, reproducibility checking, research evaluation Quality control and integrity: detecting AI-generated content, manipulation, and 'AI slop'; responsible AI assistance in authoring and reviewing Reorganizing the scientific enterprise: publication criteria, credit, careers, incentives, funding, and governance under AI-accelerated science Autonomous research agents and AI scientists Reproducibility verification and AI-generated content detection Reviewer assignment systems Position papers on the future of the scientific ecosystem (philosophy, science studies, law, policy) About the Workshop\nAI is rapidly transforming how scientific work is produced. Beyond accelerating computational workflows, AI systems increasingly support or automate major parts of research: hypothesis generation, experiment planning, analysis, writing, and even elements of wet-lab automation. As AI scientists and AI-augmented pipelines reduce the marginal cost of producing scientific artifacts, the scientific ecosystem faces a new bottleneck: evaluation, verification, and governance.\n\nWhile the production side of science has scaled dramatically, quality control (peer review, reproducibility norms, publication criteria, incentive structures, and funding allocation) is lagging behind. Record submission numbers at major AI conferences and a growing reviewer crisis make this gap visible year after year. This workshop treats the scientific enterprise itself as an object of study and intervention, and aims to establish AI for meta-science as a distinct research area at the intersection of machine learning, science policy, and the study of science. Questions we will tackle include:\n- Scaling evaluation alongside production: how can peer review, reproducibility checking, and research evaluation keep up when AI scientists can produce papers at scale, and what roles should AI reviewers and verification tools play?\n- Quality control and integrity: how do we detect and handle AI-generated content, manipulation, and \"AI slop\" in scholarly communication, while responsibly incorporating AI assistance into authoring and reviewing?\n- Reorganizing the scientific enterprise: what counts as a publication now? How should credit, careers, incentives, funding, and governance adapt to AI-accelerated science, in academia and industry alike?\n\nWho should attend\nWe bring together researchers from meta-science, philosophy of science, AI, peer review, and evaluation science with conference organizers, journal editors, and arXiv moderators. If you build AI systems for the scientific process, study the process itself, or run parts of it, this workshop is for you.\n\nFormat and outcomes\nThe program combines invited talks, contributed talks, a poster session, a panel discussion, and unconference-style breakout sessions where attendees propose their own topics for collaborative research sprints and community building. The workshop aims to produce a position paper with a set of recommendations based on participant input, to establish guidelines for ML communities on dealing with the peer-review crisis and AI in the scientific process, and to launch a lasting working group on AI for meta-science.\n\nCall for Papers\nThe workshop addresses the use, evaluation, and governance of AI systems that shape the scientific process itself: peer review, reproducibility, research evaluation, publication incentives, and funding and governance mechanisms.\n\nTwo Submission Tracks:\n- Technical Track: Methods, systems, benchmarks, datasets, and empirical analyses covering AI-assisted peer review, autonomous research agents, reproducibility verification, AI-generated content detection, and reviewer assignment systems.\n- Position Track: Argumentative papers on how the scientific ecosystem should evolve. These need not be technical and welcome perspectives from philosophy, science studies, law, and policy. Titles must begin with \"Position:\".\n\nSubmission Requirements\nSubmissions are made through the workshop's OpenReview page. Papers should use the NeurIPS 2026 LaTeX template, with the footnote changed to \"Submitted to AI for Meta-Science workshop (NeurIPS 2026)\". The NeurIPS paper checklist is not needed. Submissions are either 4 pages (concise technical contributions) or 8 pages (more substantial contributions, for example position papers), excluding references, with unlimited appendix. We welcome papers in two tracks: a technical track and a position paper track; previously published relevant work is also welcome for presentation. Novel work requires double-blind anonymization; previously published work does not. Authors may briefly disclose AI's role in producing the submission.\n\nReview Process\nSubmissions receive minimum two reviews (three preferred). Reviewers must read full main text and may use AI assistance. Decisions fall into three categories: reject, accept (poster), or accept (oral). The workshop is explicitly non-archival, allowing arXiv posting before or after. Dual submission with other venues is permitted, and previously published work is welcome.\n\nImportant Dates (Anywhere on Earth)\n- Submissions open: July 27, 2026\n- Paper submission deadline: August 29, 2026\n- Accept/reject notification: September 29, 2026\n- Camera-ready deadline: November 2026 (exact date TBA)\n- Workshop: December 2026 (exact day TBA) at NeurIPS 2026, Paris" }, { "key": "STODY", "title": "NeurIPS 2026 Workshop on AI for Stochastic Dynamics: From Theoretical Foundations to Scientific Applications", "subtitle": "NeurIPS 2026 Workshop STODY", "summary": "A NeurIPS 2026 workshop at the interface of stochastic dynamics and machine learning, convening researchers across ML, stochastic analysis, numerical simulation, and AI for science to build unified foundations, reliable algorithms, and systematic evaluation for learning stochastic dynamics in scientific applications.", "cfp_full": "Overview - Where stochastic theory meets learned dynamics.\n\nStochastic dynamics is a fundamental language for systems whose evolution is shaped by intrinsic randomness - turbulence, climate, finance, and cellular biology. Here, randomness is not noise to be removed, but an essential part of the dynamics that shapes uncertainty, long-time behavior, rare events, and regime transitions.\n\nThe same processes now sit at the heart of machine learning: diffusion and score-based generative models are built on reverse-time stochastic dynamics, neural SDEs model irregular time series, and neural operators are emerging for stochastic PDEs. Yet methods for learning stochastic dynamics still lack unified foundations, reliable algorithms, and systematic evaluation. This workshop convenes researchers across machine learning, stochastic analysis, numerical simulation, and AI for science - with particular attention to how these methods translate into scientific applications across physics, biology, chemistry, materials, and beyond.\n\nQuestion 1: How can the mathematical foundations of stochastic dynamics guide the design, analysis, and validation of machine learning systems?\n\nQuestion 2: How can stochastic structure be built into models so that learned simulators stay stable, faithful, and reliable outside the training regime?\n\nTopics of Interest\n\nScope: We focus on concrete problems at the interface of stochastic foundations and learning methodology: defining suitable learning targets, embedding stochastic structure into models and solvers, and assessing whether learned models reproduce the behavior an application requires. Topics include, but are not limited to:\n- Stochastic analysis, applied probability, and stochastic control\n- SDEs, SPDEs, and random dynamical systems\n- Neural operator methods\n- Diffusion, flow-based, and score-based generative models\n- Probabilistic forecasting and uncertainty quantification\n- Mathematical finance and stochastic modelling in climate\n- Applications in scientific discovery - physics, biology, chemistry, materials science, and molecular science\n\nCall for Papers - Submit your work\n\nAll submissions are handled through OpenReview and receive at least three reviews. We welcome two formats - page limits exclude references and appendices.\n\nSubmission format: Please use the default NeurIPS 2026 paper format for initial submissions.\n- Regular papers (up to 8 pages): Mature theoretical, algorithmic, empirical, or application-oriented contributions at the interface of stochastic dynamics and machine learning.\n- Short papers (up to 4 pages): Preliminary results, focused technical observations, negative findings, position statements, or open problems relevant to the workshop themes.\n\nImportant Dates (All deadlines are 23:59 Anywhere on Earth (AOE))\n- Aug 29: Paper submission deadline - Regular and short papers via OpenReview.\n- Aug 30 - Sep 18: Review period - Each submission receives at least three reviews.\n- Sep 19 - 28: Discussion & decision period - Reviewer discussion and final decisions.\n- Sep 29: Decision notification - Acceptance decisions sent to authors.\n- Oct 9: Camera-ready deadline - Final versions of accepted papers.\n- Dec 2026: Workshop day - Sydney - One-day in-person event during NeurIPS 2026.", "cfp_status": "published", "topics": [ "Stochastic analysis, applied probability, and stochastic control", "SDEs, SPDEs, and random dynamical systems", "Neural operator methods", "Diffusion, flow-based, and score-based generative models", "Probabilistic forecasting and uncertainty quantification", "Mathematical finance and stochastic modelling in climate", "Applications in scientific discovery - physics, biology, chemistry, materials science, and molecular science" ], "important_dates": [ { "label": "Paper submission deadline", "date": "2026-08-29" }, { "label": "Review period", "date": "2026-08-30 to 2026-09-18" }, { "label": "Discussion & decision period", "date": "2026-09-19 to 2026-09-28" }, { "label": "Decision notification", "date": "2026-09-29" }, { "label": "Camera-ready deadline", "date": "2026-10-09" }, { "label": "Workshop day", "date": "December 2026" } ], "organizers": [ "Dai Shi (University of Cambridge)", "Andi Han (University of Sydney)", "Bingxin Zhou (Shanghai Jiao Tong University)", "Junbin Gao (University of Sydney)", "Valentin De Bortoli (Google DeepMind / ENS Paris)", "Luke Thompson (University of Sydney)", "José Miguel Hernández-Lobato (University of Cambridge)" ], "speakers": [ "Nikola Kovachki (NVIDIA)", "Gary Froyland (UNSW Sydney)", "Hao Ni (UCL)", "Liang Hong (Shanghai Jiao Tong University)", "Zongyi Li (MIT/NYU)", "Anima Anandkumar (Caltech)" ], "host_url": "https://eethanshi.github.io/stochastic-dynamics-2026/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/STODY", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/STODY", "location": "Sydney, Australia", "city": "Sydney", "workshop_date": "2026-12-10", "contact": "stody.workshop@gmail.com", "tracks": [ { "key": "STODY", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/STODY", "submission_dates_raw": "" } ], "group": "science", "group_label": "AI for Science", "corpus": "NeurIPS 2026 Workshop on AI for Stochastic Dynamics: From Theoretical Foundations to Scientific Applications NeurIPS 2026 Workshop STODY A NeurIPS 2026 workshop at the interface of stochastic dynamics and machine learning, convening researchers across ML, stochastic analysis, numerical simulation, and AI for science to build unified foundations, reliable algorithms, and systematic evaluation for learning stochastic dynamics in scientific applications. Stochastic analysis, applied probability, and stochastic control SDEs, SPDEs, and random dynamical systems Neural operator methods Diffusion, flow-based, and score-based generative models Probabilistic forecasting and uncertainty quantification Mathematical finance and stochastic modelling in climate Applications in scientific discovery - physics, biology, chemistry, materials science, and molecular science Overview - Where stochastic theory meets learned dynamics.\n\nStochastic dynamics is a fundamental language for systems whose evolution is shaped by intrinsic randomness - turbulence, climate, finance, and cellular biology. Here, randomness is not noise to be removed, but an essential part of the dynamics that shapes uncertainty, long-time behavior, rare events, and regime transitions.\n\nThe same processes now sit at the heart of machine learning: diffusion and score-based generative models are built on reverse-time stochastic dynamics, neural SDEs model irregular time series, and neural operators are emerging for stochastic PDEs. Yet methods for learning stochastic dynamics still lack unified foundations, reliable algorithms, and systematic evaluation. This workshop convenes researchers across machine learning, stochastic analysis, numerical simulation, and AI for science - with particular attention to how these methods translate into scientific applications across physics, biology, chemistry, materials, and beyond.\n\nQuestion 1: How can the mathematical foundations of stochastic dynamics guide the design, analysis, and validation of machine learning systems?\n\nQuestion 2: How can stochastic structure be built into models so that learned simulators stay stable, faithful, and reliable outside the training regime?\n\nTopics of Interest\n\nScope: We focus on concrete problems at the interface of stochastic foundations and learning methodology: defining suitable learning targets, embedding stochastic structure into models and solvers, and assessing whether learned models reproduce the behavior an application requires. Topics include, but are not limited to:\n- Stochastic analysis, applied probability, and stochastic control\n- SDEs, SPDEs, and random dynamical systems\n- Neural operator methods\n- Diffusion, flow-based, and score-based generative models\n- Probabilistic forecasting and uncertainty quantification\n- Mathematical finance and stochastic modelling in climate\n- Applications in scientific discovery - physics, biology, chemistry, materials science, and molecular science\n\nCall for Papers - Submit your work\n\nAll submissions are handled through OpenReview and receive at least three reviews. We welcome two formats - page limits exclude references and appendices.\n\nSubmission format: Please use the default NeurIPS 2026 paper format for initial submissions.\n- Regular papers (up to 8 pages): Mature theoretical, algorithmic, empirical, or application-oriented contributions at the interface of stochastic dynamics and machine learning.\n- Short papers (up to 4 pages): Preliminary results, focused technical observations, negative findings, position statements, or open problems relevant to the workshop themes.\n\nImportant Dates (All deadlines are 23:59 Anywhere on Earth (AOE))\n- Aug 29: Paper submission deadline - Regular and short papers via OpenReview.\n- Aug 30 - Sep 18: Review period - Each submission receives at least three reviews.\n- Sep 19 - 28: Discussion & decision period - Reviewer discussion and final decisions.\n- Sep 29: Decision notification - Acceptance decisions sent to authors.\n- Oct 9: Camera-ready deadline - Final versions of accepted papers.\n- Dec 2026: Workshop day - Sydney - One-day in-person event during NeurIPS 2026." }, { "key": "VERICODEGEN", "title": "NeurIPS 2026 Workshop on AI for Verifiable Coding", "subtitle": "NeurIPS 2026 Workshop VERICODEGEN", "summary": "A workshop on verifiable code generation, where human-aligned agents collaborate with proof assistants, model checkers, SAT/SMT solvers, and static analyzers to co-design specifications, code, proofs, and heuristics with machine-checkable guarantees.", "cfp_full": "VeriCodeGen (NeurIPS'26)\n\nDescription\nLLM code assistants today are powerful but fundamentally untrustworthy: they hallucinate logic, mis-handle corner cases, and provide no guarantees. This workshop centers a different goal: verifiable code generation, where human-aligned agents collaborate with proof assistants, model checkers, SAT/SMT solvers, and static analyzers to co-design specifications, code, proofs, and heuristics with machine-checkable guarantees.\n\nWe argue that such agents must be built very differently from generic code LLMs. Verifiable generation requires rich specification languages (from contracts and types to temporal logics and proof terms), symbolic backends, and structured interaction loops where models propose artifacts and tools refute, repair, or certify them. Beyond code (Lean, Rust/Verus, Coq, Isabelle, Dafny, etc.), these agents must support autoformalization of informal requirements and mathematics, proof search for complex verification conditions, and heuristic discovery, e.g., invariants, lemma libraries, and solver strategies, that amplify classical formal-methods tools.\n\nWe will highlight opportunities and hard problems unique to this setting: aligning learned models with human intent expressed as specifications and proofs; orchestrating multi-agent toolchains that mix neural search with symbolic reasoning; scaling autoformalization beyond toy examples; designing training signals from compiler, verifier, or theorem-prover feedback; and building evaluation protocols that track end-to-end guarantees, not just pass@k. The goal is a roadmap toward collaborative agents that help humans understand, control, and verify the software and systems they generate.\n\nTopics\nThis workshop brings together researchers in LLMs, formal methods, programming languages, and HCI to discuss questions including (but not limited to):\n- Foundations and Architectures: Neuro-symbolic loops; verifier-in-the-loop training; multi-agent systems for code + proof.\n- Autoformalization and Specifications: From natural language and examples to contracts, types, and proof obligations; correctness and faithfulness criteria.\n- Proof, Search, and Inference: LLM-guided interactive theorem proving; certified repair; verified refactoring and migration.\n- Heuristics and Invariants: Learning loop invariants, lemmas, and solver strategies; optimizing verification backends with LLM-designed heuristics.\n- Human Alignment and UX: Interfaces for mixed-initiative specification and proof development; explanations, debugging, and trust calibration.\n- Benchmarks, Safety, and Applications: Datasets, metrics, and case studies in critical domains (infrastructure, hardware, autonomy, scientific computing).\n\nCall for Papers\nWe provide more submission details: Guidance for VeriCodeGen CFP at NeurIPS 2026.\nOpenReview submission portal: https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/VERICODEGEN\n\nTentative important dates (AoE time):\nAbstract Submission Deadline: September 8, 2026\nPaper Submission Deadline: September 10, 2026\nReview Deadline: September 26, 2026\nAcceptance/Rejection Notification Date: September 29, 2026\nCamera-Ready Submission: October 14, 2026\nWorkshop Date: December 11/12 (Sydney), or December 12/13 (Paris, Atlanta)\n\nChallenge\nCan agents make Lean proofs better, not just correct?\nThe Lean Refactor Arena is a challenge in verified proof refactoring. Participants build agents that improve existing Lean 4 proofs while preserving their theorem statements and correctness.\n- Proof size: Reduce proof-source token count compared with the reference proof.\n- Elaboration efficiency: Reduce the computational effort Lean uses to elaborate the proof.\n- Zero-shot version transfer: Keep the same proof compiling across newer Lean toolchains.\nThe benchmark brings together a range of Lean theorem-proving tasks. Submit refactored proofs through the Arena to join the public leaderboard.", "cfp_status": "published", "topics": [ "Foundations and Architectures: neuro-symbolic loops, verifier-in-the-loop training, multi-agent systems for code + proof", "Autoformalization and Specifications: from natural language to contracts, types, and proof obligations", "Proof, Search, and Inference: LLM-guided interactive theorem proving, certified repair, verified refactoring and migration", "Heuristics and Invariants: learning loop invariants, lemmas, and solver strategies", "Human Alignment and UX: interfaces for mixed-initiative specification and proof development", "Benchmarks, Safety, and Applications: datasets, metrics, and case studies in critical domains" ], "important_dates": [ { "label": "Abstract Submission Deadline", "date": "2026-09-08" }, { "label": "Paper Submission Deadline", "date": "2026-09-10" }, { "label": "Review Deadline", "date": "2026-09-26" }, { "label": "Acceptance/Rejection Notification", "date": "2026-09-29" }, { "label": "Camera-Ready Submission", "date": "2026-10-14" }, { "label": "Workshop Date", "date": "2026-12-11" } ], "organizers": [ "Wuyang Chen (Assistant Professor, Simon Fraser University)", "Soonho Kong (Principal Applied Scientist, Amazon Web Services)", "Hakjoo Oh (Professor, Korea University)", "Jingxuan He (Postdoc, University of California at Berkeley)", "Jacqueline Mitchell (Ph.D. Student, University of Southern California)", "Amanda Liu (Ph.D. Student, Massachusetts Institute of Technology)", "Zhe Ye (Ph.D. Student, University of California at Berkeley)", "Xiaodong Liu (Senior Principal Researcher and Research Manager, Microsoft)", "Varun Pant (Competition Chair, Amazon Web Services)", "Simon Frieder (Competition Chair, University of Oxford, AIMO)", "Jialin (Mike) Lu (Competition Chair, Simon Fraser University)" ], "speakers": [ "Leonardo de Moura (Amazon Web Services, Automated Reasoning Group)", "Emily First (Rutgers University)", "Vijay Ganesh (Georgia Institute of Technology)", "Patrick Li (Harmonic)", "Shan Lu (University of Chicago)", "Baishakhi Ray (Columbia University)", "Bartley Richardson (CrowdStrike)", "Shubho Sengupta (CTO, Axiom Math)", "Patrick Shafto (DARPA & Rutgers University)", "Dawn Song (University of California, Berkeley)" ], "host_url": "https://vericodegen.github.io/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/VERICODEGEN", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/VERICODEGEN", "location": "Atlanta, Georgia, United States", "city": "Atlanta", "workshop_date": "", "contact": "wuyang@sfu.ca", "tracks": [ { "key": "VERICODEGEN", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/VERICODEGEN", "submission_dates_raw": "Submission Start: Jul 14 2026 12:00AM UTC-0, Abstract Registration: Aug 27 2026 11:59AM UTC-0, Submission Deadline: Aug 29 2026 11:59AM UTC-0" } ], "group": "trust", "group_label": "Trustworthy & Safe AI", "corpus": "NeurIPS 2026 Workshop on AI for Verifiable Coding NeurIPS 2026 Workshop VERICODEGEN A workshop on verifiable code generation, where human-aligned agents collaborate with proof assistants, model checkers, SAT/SMT solvers, and static analyzers to co-design specifications, code, proofs, and heuristics with machine-checkable guarantees. Foundations and Architectures: neuro-symbolic loops, verifier-in-the-loop training, multi-agent systems for code + proof Autoformalization and Specifications: from natural language to contracts, types, and proof obligations Proof, Search, and Inference: LLM-guided interactive theorem proving, certified repair, verified refactoring and migration Heuristics and Invariants: learning loop invariants, lemmas, and solver strategies Human Alignment and UX: interfaces for mixed-initiative specification and proof development Benchmarks, Safety, and Applications: datasets, metrics, and case studies in critical domains VeriCodeGen (NeurIPS'26)\n\nDescription\nLLM code assistants today are powerful but fundamentally untrustworthy: they hallucinate logic, mis-handle corner cases, and provide no guarantees. This workshop centers a different goal: verifiable code generation, where human-aligned agents collaborate with proof assistants, model checkers, SAT/SMT solvers, and static analyzers to co-design specifications, code, proofs, and heuristics with machine-checkable guarantees.\n\nWe argue that such agents must be built very differently from generic code LLMs. Verifiable generation requires rich specification languages (from contracts and types to temporal logics and proof terms), symbolic backends, and structured interaction loops where models propose artifacts and tools refute, repair, or certify them. Beyond code (Lean, Rust/Verus, Coq, Isabelle, Dafny, etc.), these agents must support autoformalization of informal requirements and mathematics, proof search for complex verification conditions, and heuristic discovery, e.g., invariants, lemma libraries, and solver strategies, that amplify classical formal-methods tools.\n\nWe will highlight opportunities and hard problems unique to this setting: aligning learned models with human intent expressed as specifications and proofs; orchestrating multi-agent toolchains that mix neural search with symbolic reasoning; scaling autoformalization beyond toy examples; designing training signals from compiler, verifier, or theorem-prover feedback; and building evaluation protocols that track end-to-end guarantees, not just pass@k. The goal is a roadmap toward collaborative agents that help humans understand, control, and verify the software and systems they generate.\n\nTopics\nThis workshop brings together researchers in LLMs, formal methods, programming languages, and HCI to discuss questions including (but not limited to):\n- Foundations and Architectures: Neuro-symbolic loops; verifier-in-the-loop training; multi-agent systems for code + proof.\n- Autoformalization and Specifications: From natural language and examples to contracts, types, and proof obligations; correctness and faithfulness criteria.\n- Proof, Search, and Inference: LLM-guided interactive theorem proving; certified repair; verified refactoring and migration.\n- Heuristics and Invariants: Learning loop invariants, lemmas, and solver strategies; optimizing verification backends with LLM-designed heuristics.\n- Human Alignment and UX: Interfaces for mixed-initiative specification and proof development; explanations, debugging, and trust calibration.\n- Benchmarks, Safety, and Applications: Datasets, metrics, and case studies in critical domains (infrastructure, hardware, autonomy, scientific computing).\n\nCall for Papers\nWe provide more submission details: Guidance for VeriCodeGen CFP at NeurIPS 2026.\nOpenReview submission portal: https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/VERICODEGEN\n\nTentative important dates (AoE time):\nAbstract Submission Deadline: September 8, 2026\nPaper Submission Deadline: September 10, 2026\nReview Deadline: September 26, 2026\nAcceptance/Rejection Notification Date: September 29, 2026\nCamera-Ready Submission: October 14, 2026\nWorkshop Date: December 11/12 (Sydney), or December 12/13 (Paris, Atlanta)\n\nChallenge\nCan agents make Lean proofs better, not just correct?\nThe Lean Refactor Arena is a challenge in verified proof refactoring. Participants build agents that improve existing Lean 4 proofs while preserving their theorem statements and correctness.\n- Proof size: Reduce proof-source token count compared with the reference proof.\n- Elaboration efficiency: Reduce the computational effort Lean uses to elaborate the proof.\n- Zero-shot version transfer: Keep the same proof compiling across newer Lean toolchains.\nThe benchmark brings together a range of Lean theorem-proving tasks. Submit refactored proofs through the Arena to join the public leaderboard." }, { "key": "AI-Native_Academia", "title": "NeurIPS 2026 Workshop on AI-Native Academia: Authorship, Peer Review, and Conference Governance under AI", "subtitle": "AI-Native Academia @ NeurIPS 2026", "summary": "A workshop convening PC chairs of major ML venues to redesign authorship, peer review, citation, and conference governance for an academic pipeline in which LLMs and AI agents are active participants.", "cfp_full": "AI-Native Academia @ NeurIPS 2026\nAI-Native Academia: Authorship, Peer Review, and Conference Governance under AI\nNeurIPS 2026 Workshop · Atlanta, Dec 2026\n\nAcademic publishing is becoming AI-native before its governance is ready. This workshop convenes the PC chairs of ICLR, ICML, and CVPR to redesign authorship, peer review, citation, and conference governance under AI.\n\nOverview\nLLMs and AI agents are no longer just objects of study; they are now participants in the academic pipeline. Authors draft with them, reviewers summarize with them, program chairs deploy them for triage and meta-review. The institutions of science were not designed for this.\n\nFrom AI for science to AI-native academia. Most AI-for-science venues ask how AI accelerates discovery. We ask a different question: how should the institutions of science themselves change when AI is embedded in authorship, review, citation, and scholarly memory?\n\nFailure modes are appearing at every node: fabricated citations at submission, AI-generated reviews, prompt injection in manuscripts, recursive feedback into training corpora. Human-AI co-hallucination (claims no model or human would produce alone) is one among several. The workshop builds a shared taxonomy, security model, and governance framework.\n\nProgram chairs are already running scattered interventions, reviewer AI-use policies, self-ranking mechanisms, in-house detectors, adversarial-submission audits, AI-author pilots, citation audits, with no shared framework.\n\nCall for Papers\nWe invite technical, empirical, and policy submissions on nine topics organized around the AI-native academic pipeline: author -> submission -> reviewer -> AC/SAC/PC -> platform -> dissemination.\n\n1. AI-Assisted Authorship and Submission Integrity (author / submitter)\nLLM use in drafting, citation, and rebuttals; disclosure, AI-shaped novelty, fabricated claims, submission inflation, paper mills, accessibility for non-native writers, and the line between help and distortion.\n\n2. Adversarial Manuscripts and Submission-Side Attacks (submission artifact)\nHidden prompt injection, invisible text, adversarial figures, and jailbreaks against AI reviewers and citation checkers; document sanitization, submission-portal defenses, and review-tool security.\n\n3. AI-Assisted Peer Review and Reviewer Accountability (reviewer)\nAI-generated and AI-polished reviews, confidentiality risks, reviewer over-reliance, and review hallucination; measuring review quality and supporting reviewers without replacing human accountability.\n\n4. Meta-Review, PCs, and Conference-Scale Decision Support (AC / SAC / PC / organizer)\nAI for reviewer assignment, desk-reject triage, score aggregation, and collusion detection; mechanism design under AI pressure, including author self-ranking, and audit trails for AI-influenced decisions.\n\n5. Human-AI Co-Hallucination and Cross-Role Error Propagation (cross-cutting failure mode)\nTaxonomies and measurements of false claims that arise through human-AI interaction and propagate through review, citation, and reuse, with intervention points before they become institutionalized.\n\n6. Citation, Credit, and Scholarly Knowledge Integrity (citation / indexing node)\nCitation hallucination, support-of-claim verification, citation manipulation, missing-credit detection, and knowledge-graph contamination; verifying not just that a citation exists but that it supports the claim.\n\n7. Provenance, Plagiarism, Detection, and Due Process (integrity enforcement layer)\nAI-generated paper detection, watermarking, text and figure provenance, and memorization risks; detector reliability, false-positive harms to non-native authors, and appeal mechanisms.\n\n8. Recursive Scholarly Feedback Loops and Corpus Contamination (post-publication / future corpora)\nHow AI-generated papers, reviews, and citations re-enter training corpora: model collapse in scientific text, self-reinforcing citation errors, benchmark leakage, and long-term homogenization.\n\n9. Publication Infrastructure, Policy, and Venue Repositioning (platform / institution / governance)\nEnforceable AI-use policies, audit trails, confidentiality, and platform infrastructure (OpenReview, arXiv, ACM, IEEE); scaling review past 30,000+ submissions without collapsing into AI-only evaluation.\n\nSubmission tracks (via OpenReview, non-archival): Short papers (4 pages) and long papers (9 pages), references excluded. Work already accepted at NeurIPS main or other ML venues is excluded.\n\nSubmission template: Use the official NeurIPS 2026 LaTeX template on Overleaf.\n\nProspective workshop pilot\nHumanly: configurable, traceable human-AI writing. We are considering Humanly as a provenance pilot: a configurable writing environment that records in-platform edits, clipboard activity, and AI use, then issues a signed, verifiable certificate with authorship statistics and anomaly signals. The signals support human review, not an automatic verdict.\n\nThe All-PC-Chair Panel\nModerator: Atlas Wang (UT Austin / XTX Markets), framing the panel around how AI text re-entering the scientific corpus degrades future models.\nPanel theme: \"Redesigning AI Venues Under AI\" (60-min All-PC-Chair Panel + 30-min Open Q&A with the PC chairs).\nPanel question: \"By NeurIPS 2027, what must AI venues do, operationally, about human-AI co-hallucination across authorship, review, citation, and governance, including AI-review identification, prompt-injection defense, citation grounding, disclosure, and review scaling under 30k+ submissions?\"\n\nContacts: dzhang42@stevens.edu and jianing.zhu@austin.utexas.edu", "cfp_status": "published", "topics": [ "AI-Assisted Authorship and Submission Integrity", "Adversarial Manuscripts and Submission-Side Attacks", "AI-Assisted Peer Review and Reviewer Accountability", "Meta-Review, PCs, and Conference-Scale Decision Support", "Human-AI Co-Hallucination and Cross-Role Error Propagation", "Citation, Credit, and Scholarly Knowledge Integrity", "Provenance, Plagiarism, Detection, and Due Process", "Recursive Scholarly Feedback Loops and Corpus Contamination", "Publication Infrastructure, Policy, and Venue Repositioning" ], "important_dates": [ { "label": "Call for Papers opens", "date": "2026-07-18" }, { "label": "Submission deadline", "date": "2026-08-29" }, { "label": "Reviews", "date": "2026-09-01 to 2026-09-26" }, { "label": "Author notification", "date": "2026-09-29" }, { "label": "Camera-ready", "date": "2026-10-10" }, { "label": "Workshop date", "date": "2026-12-12 to 2026-12-13" } ], "organizers": [ "Denghui Zhang (Stevens Institute of Technology)", "Jianing Zhu (UT Austin)", "Gopal Ramchurn (University of Southampton)", "Manling Li (Northwestern)", "Atlas Wang (UT Austin / XTX Markets)" ], "speakers": [ "James Zou (Stanford)", "Kyunghyun Cho (NYU)", "Hima Lakkaraju (Harvard / Google)", "Hiromu Yakura (MPI / Anthropic)", "Yian Yin (Cornell University)", "Lin Peng (Baruch / CUNY)", "Nihar B. Shah (CMU)", "Yisong Yue (Caltech)", "Sharon Li (UW-Madison)", "Mert Sabuncu (Cornell Tech / Weill Cornell)", "Humphrey Shi (Georgia Tech / NVIDIA)" ], "host_url": "https://ai-native-academia.github.io", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AI-Native_Academia", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AI-Native_Academia", "location": "Atlanta", "city": "Atlanta", "workshop_date": "2026-12-12", "contact": "dzhang42@stevens.edu", "tracks": [ { "key": "AI-Native_Academia", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AI-Native_Academia", "submission_dates_raw": "" } ], "group": "society", "group_label": "Society, Applications & Community", "corpus": "NeurIPS 2026 Workshop on AI-Native Academia: Authorship, Peer Review, and Conference Governance under AI AI-Native Academia @ NeurIPS 2026 A workshop convening PC chairs of major ML venues to redesign authorship, peer review, citation, and conference governance for an academic pipeline in which LLMs and AI agents are active participants. AI-Assisted Authorship and Submission Integrity Adversarial Manuscripts and Submission-Side Attacks AI-Assisted Peer Review and Reviewer Accountability Meta-Review, PCs, and Conference-Scale Decision Support Human-AI Co-Hallucination and Cross-Role Error Propagation Citation, Credit, and Scholarly Knowledge Integrity Provenance, Plagiarism, Detection, and Due Process Recursive Scholarly Feedback Loops and Corpus Contamination Publication Infrastructure, Policy, and Venue Repositioning AI-Native Academia @ NeurIPS 2026\nAI-Native Academia: Authorship, Peer Review, and Conference Governance under AI\nNeurIPS 2026 Workshop · Atlanta, Dec 2026\n\nAcademic publishing is becoming AI-native before its governance is ready. This workshop convenes the PC chairs of ICLR, ICML, and CVPR to redesign authorship, peer review, citation, and conference governance under AI.\n\nOverview\nLLMs and AI agents are no longer just objects of study; they are now participants in the academic pipeline. Authors draft with them, reviewers summarize with them, program chairs deploy them for triage and meta-review. The institutions of science were not designed for this.\n\nFrom AI for science to AI-native academia. Most AI-for-science venues ask how AI accelerates discovery. We ask a different question: how should the institutions of science themselves change when AI is embedded in authorship, review, citation, and scholarly memory?\n\nFailure modes are appearing at every node: fabricated citations at submission, AI-generated reviews, prompt injection in manuscripts, recursive feedback into training corpora. Human-AI co-hallucination (claims no model or human would produce alone) is one among several. The workshop builds a shared taxonomy, security model, and governance framework.\n\nProgram chairs are already running scattered interventions, reviewer AI-use policies, self-ranking mechanisms, in-house detectors, adversarial-submission audits, AI-author pilots, citation audits, with no shared framework.\n\nCall for Papers\nWe invite technical, empirical, and policy submissions on nine topics organized around the AI-native academic pipeline: author -> submission -> reviewer -> AC/SAC/PC -> platform -> dissemination.\n\n1. AI-Assisted Authorship and Submission Integrity (author / submitter)\nLLM use in drafting, citation, and rebuttals; disclosure, AI-shaped novelty, fabricated claims, submission inflation, paper mills, accessibility for non-native writers, and the line between help and distortion.\n\n2. Adversarial Manuscripts and Submission-Side Attacks (submission artifact)\nHidden prompt injection, invisible text, adversarial figures, and jailbreaks against AI reviewers and citation checkers; document sanitization, submission-portal defenses, and review-tool security.\n\n3. AI-Assisted Peer Review and Reviewer Accountability (reviewer)\nAI-generated and AI-polished reviews, confidentiality risks, reviewer over-reliance, and review hallucination; measuring review quality and supporting reviewers without replacing human accountability.\n\n4. Meta-Review, PCs, and Conference-Scale Decision Support (AC / SAC / PC / organizer)\nAI for reviewer assignment, desk-reject triage, score aggregation, and collusion detection; mechanism design under AI pressure, including author self-ranking, and audit trails for AI-influenced decisions.\n\n5. Human-AI Co-Hallucination and Cross-Role Error Propagation (cross-cutting failure mode)\nTaxonomies and measurements of false claims that arise through human-AI interaction and propagate through review, citation, and reuse, with intervention points before they become institutionalized.\n\n6. Citation, Credit, and Scholarly Knowledge Integrity (citation / indexing node)\nCitation hallucination, support-of-claim verification, citation manipulation, missing-credit detection, and knowledge-graph contamination; verifying not just that a citation exists but that it supports the claim.\n\n7. Provenance, Plagiarism, Detection, and Due Process (integrity enforcement layer)\nAI-generated paper detection, watermarking, text and figure provenance, and memorization risks; detector reliability, false-positive harms to non-native authors, and appeal mechanisms.\n\n8. Recursive Scholarly Feedback Loops and Corpus Contamination (post-publication / future corpora)\nHow AI-generated papers, reviews, and citations re-enter training corpora: model collapse in scientific text, self-reinforcing citation errors, benchmark leakage, and long-term homogenization.\n\n9. Publication Infrastructure, Policy, and Venue Repositioning (platform / institution / governance)\nEnforceable AI-use policies, audit trails, confidentiality, and platform infrastructure (OpenReview, arXiv, ACM, IEEE); scaling review past 30,000+ submissions without collapsing into AI-only evaluation.\n\nSubmission tracks (via OpenReview, non-archival): Short papers (4 pages) and long papers (9 pages), references excluded. Work already accepted at NeurIPS main or other ML venues is excluded.\n\nSubmission template: Use the official NeurIPS 2026 LaTeX template on Overleaf.\n\nProspective workshop pilot\nHumanly: configurable, traceable human-AI writing. We are considering Humanly as a provenance pilot: a configurable writing environment that records in-platform edits, clipboard activity, and AI use, then issues a signed, verifiable certificate with authorship statistics and anomaly signals. The signals support human review, not an automatic verdict.\n\nThe All-PC-Chair Panel\nModerator: Atlas Wang (UT Austin / XTX Markets), framing the panel around how AI text re-entering the scientific corpus degrades future models.\nPanel theme: \"Redesigning AI Venues Under AI\" (60-min All-PC-Chair Panel + 30-min Open Q&A with the PC chairs).\nPanel question: \"By NeurIPS 2027, what must AI venues do, operationally, about human-AI co-hallucination across authorship, review, citation, and governance, including AI-review identification, prompt-injection defense, citation grounding, disclosure, and review scaling under 30k+ submissions?\"\n\nContacts: dzhang42@stevens.edu and jianing.zhu@austin.utexas.edu" }, { "key": "Child_Safety_in_AI", "title": "NeurIPS 2026 Workshop on Child Safety in AI", "subtitle": "Child Safety in AI Workshop", "summary": "A workshop bringing together researchers, practitioners, and policymakers to address the technical and sociotechnical challenges of protecting children in AI ecosystems, treating child safety as a core dimension of AI safety.", "cfp_full": "About\n\nModern AI systems introduce new risks to children, including developmental harms such as increased reliance on AI and unhealthy social or emotional attachment; interactions that may contribute to self-harm or suicidal ideation; and the creation or misuse of synthetic content for harassment, grooming, extortion, and sexual abuse.\n\nAt the same time, child safety places unusual constraints on conventional AI safety research. Harmful data may be illegal or unethical to access, real-world evaluations are limited, and effective interventions must consider broad ecosystem effects and often require cross-sector collaboration with groups such as NGOs, hotlines, law enforcement, regulators, and child-protection experts.\n\nThe workshop considers child safety as a core dimension of AI safety, and as a critical test case for the development, deployment, maintenance, and governance of safe AI systems.\n\nResearch directions\n\nSafe Data, Evaluation, and Benchmarking\n- Safe data curation and governance for child-related data\n- Evaluation methodologies under restricted access settings (e.g., proxy tasks, synthetic benchmarks)\n- Measurement and benchmarking of child safety risks (e.g., datasets, metrics, eval protocols, long-term evals)\n- Auditing and interpretability methods for detecting unsafe capabilities\n- Data-free or privacy-preserving auditing techniques\n\nRobust and Safe Model Design\n- Preventing the emergence of harmful capabilities (e.g., defenses against jailbreaking, resilience to harmful fine-tuning, preventing concept fusion, data cleaning)\n- Evaluating capability degradation when implementing safety solutions\n- Adversarial robustness and red teaming for child safety\n- Machine unlearning and concept erasure with strong guarantees\n- Child safety in multimodal and agentic AI systems\n\nDeployment, Monitoring, and Ecosystem Safeguards\n- Robust safeguards for model deployment (e.g., input/output filtering, monitoring, open-weight model defenses)\n- Content provenance, watermarking, and traceability\n- Safety-privacy trade-offs in AI systems\n- Protection of user-generated content from manipulation\n- Cross-platform and ecosystem-level safety coordination\n- Human factors and moderator well-being\n\nHuman-Centered Design, Policy, and Societal Implications\n- Child-centered safety design and age-appropriate interfaces\n- Policy, governance, and regulatory frameworks\n- Collaboration with external stakeholders (NGOs, law enforcement, hotlines)\n- Ethical, legal, and societal implications, including global perspectives\n\nCall for papers\n\nWe invite full papers, works-in-progress, position papers, and open-problem submissions of up to four pages, excluding references, using the NeurIPS 2026 template (with the 'dblblindworkshop' option). Submissions must be anonymous and may include an optional unlimited-length appendix.\n\nAccepted papers will be non-archival and may be submitted to other venues. Submissions will receive double-blind review, and evaluation will prioritize the potential to stimulate productive workshop discussion alongside technical soundness.\n\nSubmission deadline: August 29, 2026, AoE\nNotification: September 29, 2026, AoE\nSubmission site: OpenReview", "cfp_status": "published", "topics": [ "Safe data curation and governance for child-related data", "Evaluation methodologies under restricted access settings (proxy tasks, synthetic benchmarks)", "Measurement and benchmarking of child safety risks", "Auditing and interpretability methods for detecting unsafe capabilities", "Data-free or privacy-preserving auditing techniques", "Preventing the emergence of harmful capabilities (defenses against jailbreaking, resilience to harmful fine-tuning, preventing concept fusion, data cleaning)", "Evaluating capability degradation when implementing safety solutions", "Adversarial robustness and red teaming for child safety", "Machine unlearning and concept erasure with strong guarantees", "Child safety in multimodal and agentic AI systems", "Robust safeguards for model deployment (input/output filtering, monitoring, open-weight model defenses)", "Content provenance, watermarking, and traceability", "Safety-privacy trade-offs in AI systems", "Protection of user-generated content from manipulation", "Cross-platform and ecosystem-level safety coordination", "Human factors and moderator well-being", "Child-centered safety design and age-appropriate interfaces", "Policy, governance, and regulatory frameworks", "Collaboration with external stakeholders (NGOs, law enforcement, hotlines)", "Ethical, legal, and societal implications, including global perspectives" ], "important_dates": [ { "label": "Submission Deadline", "date": "2026-08-29" }, { "label": "Notification", "date": "2026-09-29" } ], "organizers": [ "Rebecca Portnoff", "Virginia Smith", "Andrew Strait", "Ashia Wilson", "Campbell Wilson", "Neil Kale", "Aashiq Muhamed", "Vinith Suriyakumar" ], "speakers": [ "Ana-Maria Cretu (CISPA)", "Sauvik Das (Carnegie Mellon University)", "Julie Inman Grant (Australia's eSafety Commissioner)", "Riana Pfefferkorn (Stanford University)", "Robbie Torney (Common Sense Media)", "Miranda Wei (EPFL)" ], "host_url": "https://childsafety-ai.github.io/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/Child_Safety_in_AI", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/Child_Safety_in_AI", "location": "Atlanta, GA", "city": "Atlanta", "workshop_date": "2026-12-12", "contact": "smithv@cmu.edu", "tracks": [ { "key": "Child_Safety_in_AI", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/Child_Safety_in_AI", "submission_dates_raw": "" } ], "group": "trust", "group_label": "Trustworthy & Safe AI", "corpus": "NeurIPS 2026 Workshop on Child Safety in AI Child Safety in AI Workshop A workshop bringing together researchers, practitioners, and policymakers to address the technical and sociotechnical challenges of protecting children in AI ecosystems, treating child safety as a core dimension of AI safety. Safe data curation and governance for child-related data Evaluation methodologies under restricted access settings (proxy tasks, synthetic benchmarks) Measurement and benchmarking of child safety risks Auditing and interpretability methods for detecting unsafe capabilities Data-free or privacy-preserving auditing techniques Preventing the emergence of harmful capabilities (defenses against jailbreaking, resilience to harmful fine-tuning, preventing concept fusion, data cleaning) Evaluating capability degradation when implementing safety solutions Adversarial robustness and red teaming for child safety Machine unlearning and concept erasure with strong guarantees Child safety in multimodal and agentic AI systems Robust safeguards for model deployment (input/output filtering, monitoring, open-weight model defenses) Content provenance, watermarking, and traceability Safety-privacy trade-offs in AI systems Protection of user-generated content from manipulation Cross-platform and ecosystem-level safety coordination Human factors and moderator well-being Child-centered safety design and age-appropriate interfaces Policy, governance, and regulatory frameworks Collaboration with external stakeholders (NGOs, law enforcement, hotlines) Ethical, legal, and societal implications, including global perspectives About\n\nModern AI systems introduce new risks to children, including developmental harms such as increased reliance on AI and unhealthy social or emotional attachment; interactions that may contribute to self-harm or suicidal ideation; and the creation or misuse of synthetic content for harassment, grooming, extortion, and sexual abuse.\n\nAt the same time, child safety places unusual constraints on conventional AI safety research. Harmful data may be illegal or unethical to access, real-world evaluations are limited, and effective interventions must consider broad ecosystem effects and often require cross-sector collaboration with groups such as NGOs, hotlines, law enforcement, regulators, and child-protection experts.\n\nThe workshop considers child safety as a core dimension of AI safety, and as a critical test case for the development, deployment, maintenance, and governance of safe AI systems.\n\nResearch directions\n\nSafe Data, Evaluation, and Benchmarking\n- Safe data curation and governance for child-related data\n- Evaluation methodologies under restricted access settings (e.g., proxy tasks, synthetic benchmarks)\n- Measurement and benchmarking of child safety risks (e.g., datasets, metrics, eval protocols, long-term evals)\n- Auditing and interpretability methods for detecting unsafe capabilities\n- Data-free or privacy-preserving auditing techniques\n\nRobust and Safe Model Design\n- Preventing the emergence of harmful capabilities (e.g., defenses against jailbreaking, resilience to harmful fine-tuning, preventing concept fusion, data cleaning)\n- Evaluating capability degradation when implementing safety solutions\n- Adversarial robustness and red teaming for child safety\n- Machine unlearning and concept erasure with strong guarantees\n- Child safety in multimodal and agentic AI systems\n\nDeployment, Monitoring, and Ecosystem Safeguards\n- Robust safeguards for model deployment (e.g., input/output filtering, monitoring, open-weight model defenses)\n- Content provenance, watermarking, and traceability\n- Safety-privacy trade-offs in AI systems\n- Protection of user-generated content from manipulation\n- Cross-platform and ecosystem-level safety coordination\n- Human factors and moderator well-being\n\nHuman-Centered Design, Policy, and Societal Implications\n- Child-centered safety design and age-appropriate interfaces\n- Policy, governance, and regulatory frameworks\n- Collaboration with external stakeholders (NGOs, law enforcement, hotlines)\n- Ethical, legal, and societal implications, including global perspectives\n\nCall for papers\n\nWe invite full papers, works-in-progress, position papers, and open-problem submissions of up to four pages, excluding references, using the NeurIPS 2026 template (with the 'dblblindworkshop' option). Submissions must be anonymous and may include an optional unlimited-length appendix.\n\nAccepted papers will be non-archival and may be submitted to other venues. Submissions will receive double-blind review, and evaluation will prioritize the potential to stimulate productive workshop discussion alongside technical soundness.\n\nSubmission deadline: August 29, 2026, AoE\nNotification: September 29, 2026, AoE\nSubmission site: OpenReview" }, { "key": "CLEA", "title": "NeurIPS 2026 Workshop on Continual Learning for Enterprise AI Agents", "subtitle": "NeurIPS-CLEA 2026", "summary": "A NeurIPS 2026 workshop bringing together researchers and practitioners on continual learning, agent systems, reinforcement learning, and enterprise AI deployment to discuss methodologies, benchmarks, evaluation protocols, and real-world challenges in building adaptive enterprise AI agents.", "cfp_full": "Continual learning is becoming a critical capability for enterprise AI agents operating in evolving environments. This workshop brings together researchers and practitioners working on continual learning, agent systems, reinforcement learning, and enterprise AI deployment to discuss methodologies, benchmarks, evaluation protocols, and real-world challenges in building adaptive AI agents.\n\nTopics of Interest\n\nModel Adaptation and Learning\n- Catastrophic forgetting detection and mitigation, Compute- and cost-efficiency, Task-agnostic adaptation in non-stationary environments, Personalized and role-aware continual adaptation\n\nAgent Orchestration and Workflow Evolution\n- Lifelong tool-use and workflow adaptation, Harness-level continual improvement, Multi-agent and human-agent ecosystems\n\nMemory and Organizational Knowledge\n- Memory and enterprise knowledge evolution, Feedback-driven learning from enterprise signals\n\nSystems, Governance, and Deployment\n- Compliance, safety, and alignment drift, Evaluation and harness engineering for enterprise agents, System-level continual learning and deployment\n\nApplications and Case Studies\n- Real-world continual learning deployments such as IT operations, customer support, compliance monitoring, financial services, and asset management\n\nCall for Papers\n\nImportant Dates\n- Submission Deadline: Sep 4, 2026 AOE\n- Notification Date: Sep 29, 2026 AOE\n- Workshop: Dec 12 or 13, 2026\n\nSubmission Tracks\n\nResearch Papers - 5-9 pages (excluding unlimited references and appendices)\nResearch papers should present original research contributions related to the workshop topics. Submissions should describe novel methods, systems, benchmarks, theoretical insights, or empirical studies.\n\nOpinion Papers - Up to 5 pages (excluding unlimited references)\nOpinion papers include position papers, vision papers, demonstrations, or experience reports that articulate new research directions, identify open challenges, or discuss opportunities in continual learning for enterprise AI agents. Extensive experimental evaluation is not required.\n\nWe also welcome papers based on previously published work or work currently under review, provided the submission is clearly identified as such. Such papers are encouraged when they help stimulate discussion within the workshop. Submissions describing preliminary work or ongoing research will not preclude subsequent publication in conferences or journals.\n\nSubmission and Review\n- Submit your paper through OpenReview.\n- Please follow the NeurIPS 2026 formatting guidelines.\n- All submissions will undergo a double-blind peer review process.\n\nPresentation\nAll accepted papers will be presented during the workshop poster session. A small number of outstanding papers will also be selected for spotlight oral presentations.\n\nCall for Reviewers\nIf you work on continual learning, AI agents, enterprise AI, or related areas, we'd love to hear from you! We invite experienced researchers and practitioners to join our review panel. As a reviewer, you'll help shape the quality of presentations and contribute to the success of the workshop.", "cfp_status": "published", "topics": [ "Model Adaptation and Learning (catastrophic forgetting detection/mitigation, compute- and cost-efficiency, task-agnostic adaptation in non-stationary environments, personalized and role-aware continual adaptation)", "Agent Orchestration and Workflow Evolution (lifelong tool-use and workflow adaptation, harness-level continual improvement, multi-agent and human-agent ecosystems)", "Memory and Organizational Knowledge (memory and enterprise knowledge evolution, feedback-driven learning from enterprise signals)", "Systems, Governance, and Deployment (compliance, safety, and alignment drift; evaluation and harness engineering; system-level continual learning and deployment)", "Applications and Case Studies (IT operations, customer support, compliance monitoring, financial services, asset management)" ], "important_dates": [ { "label": "Submission Deadline", "date": "2026-09-04" }, { "label": "Notification Date", "date": "2026-09-29" }, { "label": "Workshop", "date": "December 12 or 13, 2026" } ], "organizers": [ "Naoki Abe (IBM)", "Aurélie Lozano (IBM)", "Xi Yang (IBM)", "Yu Deng (IBM)", "Meng Jiang (University of Notre Dame)", "Chengxiang Zhai (University of Illinois Urbana-Champaign)", "Michael L. Littman (Brown University)" ], "speakers": [ "Azalia Mirhoseini (Stanford University)", "Bing Liu (University of Illinois Chicago)" ], "host_url": "https://clea-neurips.github.io/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/CLEA", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/CLEA", "location": "Atlanta", "city": "Atlanta", "workshop_date": "2026-12-14", "contact": "xi.yang@ibm.com", "tracks": [ { "key": "CLEA", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/CLEA", "submission_dates_raw": "" } ], "group": "agents", "group_label": "Agents & Agentic Systems", "corpus": "NeurIPS 2026 Workshop on Continual Learning for Enterprise AI Agents NeurIPS-CLEA 2026 A NeurIPS 2026 workshop bringing together researchers and practitioners on continual learning, agent systems, reinforcement learning, and enterprise AI deployment to discuss methodologies, benchmarks, evaluation protocols, and real-world challenges in building adaptive enterprise AI agents. Model Adaptation and Learning (catastrophic forgetting detection/mitigation, compute- and cost-efficiency, task-agnostic adaptation in non-stationary environments, personalized and role-aware continual adaptation) Agent Orchestration and Workflow Evolution (lifelong tool-use and workflow adaptation, harness-level continual improvement, multi-agent and human-agent ecosystems) Memory and Organizational Knowledge (memory and enterprise knowledge evolution, feedback-driven learning from enterprise signals) Systems, Governance, and Deployment (compliance, safety, and alignment drift; evaluation and harness engineering; system-level continual learning and deployment) Applications and Case Studies (IT operations, customer support, compliance monitoring, financial services, asset management) Continual learning is becoming a critical capability for enterprise AI agents operating in evolving environments. This workshop brings together researchers and practitioners working on continual learning, agent systems, reinforcement learning, and enterprise AI deployment to discuss methodologies, benchmarks, evaluation protocols, and real-world challenges in building adaptive AI agents.\n\nTopics of Interest\n\nModel Adaptation and Learning\n- Catastrophic forgetting detection and mitigation, Compute- and cost-efficiency, Task-agnostic adaptation in non-stationary environments, Personalized and role-aware continual adaptation\n\nAgent Orchestration and Workflow Evolution\n- Lifelong tool-use and workflow adaptation, Harness-level continual improvement, Multi-agent and human-agent ecosystems\n\nMemory and Organizational Knowledge\n- Memory and enterprise knowledge evolution, Feedback-driven learning from enterprise signals\n\nSystems, Governance, and Deployment\n- Compliance, safety, and alignment drift, Evaluation and harness engineering for enterprise agents, System-level continual learning and deployment\n\nApplications and Case Studies\n- Real-world continual learning deployments such as IT operations, customer support, compliance monitoring, financial services, and asset management\n\nCall for Papers\n\nImportant Dates\n- Submission Deadline: Sep 4, 2026 AOE\n- Notification Date: Sep 29, 2026 AOE\n- Workshop: Dec 12 or 13, 2026\n\nSubmission Tracks\n\nResearch Papers - 5-9 pages (excluding unlimited references and appendices)\nResearch papers should present original research contributions related to the workshop topics. Submissions should describe novel methods, systems, benchmarks, theoretical insights, or empirical studies.\n\nOpinion Papers - Up to 5 pages (excluding unlimited references)\nOpinion papers include position papers, vision papers, demonstrations, or experience reports that articulate new research directions, identify open challenges, or discuss opportunities in continual learning for enterprise AI agents. Extensive experimental evaluation is not required.\n\nWe also welcome papers based on previously published work or work currently under review, provided the submission is clearly identified as such. Such papers are encouraged when they help stimulate discussion within the workshop. Submissions describing preliminary work or ongoing research will not preclude subsequent publication in conferences or journals.\n\nSubmission and Review\n- Submit your paper through OpenReview.\n- Please follow the NeurIPS 2026 formatting guidelines.\n- All submissions will undergo a double-blind peer review process.\n\nPresentation\nAll accepted papers will be presented during the workshop poster session. A small number of outstanding papers will also be selected for spotlight oral presentations.\n\nCall for Reviewers\nIf you work on continual learning, AI agents, enterprise AI, or related areas, we'd love to hear from you! We invite experienced researchers and practitioners to join our review panel. As a reviewer, you'll help shape the quality of presentations and contribute to the success of the workshop." }, { "key": "CL4FMAgents", "title": "NeurIPS 2026 Workshop on Continual Learning in the Era of Foundation Models and Embodied Agents", "subtitle": "CL4FMAgents @ NeurIPS 2026", "summary": "CL4FMAgents focuses on continual learning as a shared challenge for foundation models and embodied agents in dynamic, open-ended, and interactive environments, bringing together researchers from continual learning, foundation model adaptation, embodied intelligence, and robot learning.", "cfp_full": "NeurIPS 2026 Workshop on Continual Learning in the Era of Foundation Models and Embodied Agents\nDecember 11-12, 2026 - Sydney, Australia\nContact: neurips26.cl4fmagents@gmail.com\n\nAbout\nThis workshop focuses on continual learning as a shared challenge for foundation models and embodied agents in dynamic, open-ended, and interactive environments. We intentionally bring these two areas together because they are becoming increasingly intertwined in modern AI: foundation models are becoming key components of embodied systems, while embodied settings expose models to non-stationary conditions that require continual adaptation.\n\nThis intersection gives rise to common challenges, including catastrophic forgetting, memory and knowledge consolidation, online adaptation, long-term skill acquisition, and safety during updates. The workshop centers specifically on continual learning problems that arise within and across these domains.\n\nThe workshop will bring together researchers from continual learning, foundation model adaptation, embodied intelligence, and robot learning to address shared challenges and identify new opportunities for cross-pollination.\n\nCall for Papers\nWe invite submissions in two tracks: regular papers (up to 8 pages excluding references) and short papers (up to 4 pages excluding references). Contributions may take the form of research papers, position or perspective pieces, benchmarks and datasets, systems and applications papers, negative or reproducibility results, or other interdisciplinary work spanning continual, lifelong, and online learning across foundation models, LLM agents, and embodied systems. We welcome work ranging from theory and algorithms to large-scale empirical studies and real-world deployments. All submissions are non-archival and will be managed through OpenReview.\n\nTopics of interest include but are not limited to:\n- Continual and lifelong pre-training, post-training, and alignment of foundation models (LLMs, VLMs, and multimodal models)\n- Continual learning, self-evolution, and long-term skill acquisition of LLM and foundation-model agents\n- Continual and lifelong learning for embodied agents, robotics, control, and world models\n- Memory architectures, knowledge consolidation, model editing, and retrieval-augmented adaptation\n- Online, test-time, and streaming adaptation under distribution shift and non-stationarity\n- Catastrophic forgetting, stability-plasticity trade-offs, and forward/backward transfer\n- Theory and foundations of continual learning: generalization, optimization, and scaling laws\n- Safety, robustness, privacy, and reliability during continual updates\n- Benchmarks, evaluation protocols, and metrics for lifelong and continually learning systems\n- Applications and deployment: scientific discovery, healthcare, autonomous systems, and personalization\n\nEach submission will receive at least three reviews. Submissions must not re-present finalized work previously published at ML venues.\nSubmission link: Submit on OpenReview\n\nImportant Dates\n- Submission Deadline: August 29, 2026 (AoE)\n- Notification: September 29, 2026 (AoE)\n- Camera-ready: October 10, 2026\n- Workshop Date: December 11-12, 2026", "cfp_status": "published", "topics": [ "Continual and lifelong pre-training, post-training, and alignment of foundation models (LLMs, VLMs, and multimodal models)", "Continual learning, self-evolution, and long-term skill acquisition of LLM and foundation-model agents", "Continual and lifelong learning for embodied agents, robotics, control, and world models", "Memory architectures, knowledge consolidation, model editing, and retrieval-augmented adaptation", "Online, test-time, and streaming adaptation under distribution shift and non-stationarity", "Catastrophic forgetting, stability-plasticity trade-offs, and forward/backward transfer", "Theory and foundations of continual learning: generalization, optimization, and scaling laws", "Safety, robustness, privacy, and reliability during continual updates", "Benchmarks, evaluation protocols, and metrics for lifelong and continually learning systems", "Applications and deployment: scientific discovery, healthcare, autonomous systems, and personalization" ], "important_dates": [ { "label": "Submission Deadline (AoE)", "date": "2026-08-29" }, { "label": "Notification (AoE)", "date": "2026-09-29" }, { "label": "Camera-ready", "date": "2026-10-10" }, { "label": "Workshop", "date": "2026-12-11" } ], "organizers": [ "Jonghyun Choi (Seoul National University)", "Xiao-Ming Wu (The Hong Kong Polytechnic University)", "Rahaf Aljundi (Toyota Motor Europe)", "Jorge Mendez-Mendez (Stony Brook University)", "Liyuan Wang (Tsinghua University)", "Yujie Feng (The Hong Kong Polytechnic University)", "Yuankai Luo (Nanjing University)" ], "speakers": [ "Irina Rish (Université de Montréal / Mila)", "Chelsea Finn (Stanford University)", "Abhishek Gupta (University of Washington)", "Gido M. van de Ven (University of Groningen)", "Jeff Clune (University of British Columbia / Vector Institute)", "Kianté Brantley (Harvard University)" ], "host_url": "https://neurips26-cl4fmagents.github.io", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/CL4FMAgents", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/CL4FMAgents", "location": "Sydney, Australia", "city": "Sydney", "workshop_date": "2026-12-11", "contact": "scenefyj@gmai.com", "tracks": [ { "key": "CL4FMAgents", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/CL4FMAgents", "submission_dates_raw": "" } ], "group": "fm", "group_label": "Foundation Models & LLMs", "corpus": "NeurIPS 2026 Workshop on Continual Learning in the Era of Foundation Models and Embodied Agents CL4FMAgents @ NeurIPS 2026 CL4FMAgents focuses on continual learning as a shared challenge for foundation models and embodied agents in dynamic, open-ended, and interactive environments, bringing together researchers from continual learning, foundation model adaptation, embodied intelligence, and robot learning. Continual and lifelong pre-training, post-training, and alignment of foundation models (LLMs, VLMs, and multimodal models) Continual learning, self-evolution, and long-term skill acquisition of LLM and foundation-model agents Continual and lifelong learning for embodied agents, robotics, control, and world models Memory architectures, knowledge consolidation, model editing, and retrieval-augmented adaptation Online, test-time, and streaming adaptation under distribution shift and non-stationarity Catastrophic forgetting, stability-plasticity trade-offs, and forward/backward transfer Theory and foundations of continual learning: generalization, optimization, and scaling laws Safety, robustness, privacy, and reliability during continual updates Benchmarks, evaluation protocols, and metrics for lifelong and continually learning systems Applications and deployment: scientific discovery, healthcare, autonomous systems, and personalization NeurIPS 2026 Workshop on Continual Learning in the Era of Foundation Models and Embodied Agents\nDecember 11-12, 2026 - Sydney, Australia\nContact: neurips26.cl4fmagents@gmail.com\n\nAbout\nThis workshop focuses on continual learning as a shared challenge for foundation models and embodied agents in dynamic, open-ended, and interactive environments. We intentionally bring these two areas together because they are becoming increasingly intertwined in modern AI: foundation models are becoming key components of embodied systems, while embodied settings expose models to non-stationary conditions that require continual adaptation.\n\nThis intersection gives rise to common challenges, including catastrophic forgetting, memory and knowledge consolidation, online adaptation, long-term skill acquisition, and safety during updates. The workshop centers specifically on continual learning problems that arise within and across these domains.\n\nThe workshop will bring together researchers from continual learning, foundation model adaptation, embodied intelligence, and robot learning to address shared challenges and identify new opportunities for cross-pollination.\n\nCall for Papers\nWe invite submissions in two tracks: regular papers (up to 8 pages excluding references) and short papers (up to 4 pages excluding references). Contributions may take the form of research papers, position or perspective pieces, benchmarks and datasets, systems and applications papers, negative or reproducibility results, or other interdisciplinary work spanning continual, lifelong, and online learning across foundation models, LLM agents, and embodied systems. We welcome work ranging from theory and algorithms to large-scale empirical studies and real-world deployments. All submissions are non-archival and will be managed through OpenReview.\n\nTopics of interest include but are not limited to:\n- Continual and lifelong pre-training, post-training, and alignment of foundation models (LLMs, VLMs, and multimodal models)\n- Continual learning, self-evolution, and long-term skill acquisition of LLM and foundation-model agents\n- Continual and lifelong learning for embodied agents, robotics, control, and world models\n- Memory architectures, knowledge consolidation, model editing, and retrieval-augmented adaptation\n- Online, test-time, and streaming adaptation under distribution shift and non-stationarity\n- Catastrophic forgetting, stability-plasticity trade-offs, and forward/backward transfer\n- Theory and foundations of continual learning: generalization, optimization, and scaling laws\n- Safety, robustness, privacy, and reliability during continual updates\n- Benchmarks, evaluation protocols, and metrics for lifelong and continually learning systems\n- Applications and deployment: scientific discovery, healthcare, autonomous systems, and personalization\n\nEach submission will receive at least three reviews. Submissions must not re-present finalized work previously published at ML venues.\nSubmission link: Submit on OpenReview\n\nImportant Dates\n- Submission Deadline: August 29, 2026 (AoE)\n- Notification: September 29, 2026 (AoE)\n- Camera-ready: October 10, 2026\n- Workshop Date: December 11-12, 2026" }, { "key": "CWM", "title": "NeurIPS 2026 Workshop on Continual World Models", "subtitle": "NeurIPS 2026 Workshop CWM", "summary": "A NeurIPS 2026 workshop bringing together computer vision, robotics, reinforcement learning, generative modeling, and cognitive science to define a research agenda for continual world models — systems that improve from the worlds they observe, imagine, and act within rather than remaining frozen predictors.", "cfp_full": "Continual World Models\n\nA world model should change when the world does.\n\nThe missing continual axis\n\nWorld models increasingly encode rich motion and physical priors, yet most remain frozen predictors: trained once, then expected to generalize to whatever appears after deployment. Open-ended environments demand something more exacting, the capacity to notice prediction failures, absorb new evidence, and revise an evolving model of the world.\n\nThis workshop brings together computer vision, robotics, reinforcement learning, generative modeling, and cognitive science to define a research agenda for continual world models. We are interested in systems that do not merely extrapolate from fixed datasets, but improve from the worlds they observe, imagine, and act within.\n\nCall for papers\n\nWe invite 4-8 page non-archival submissions that advance our understanding of continual improvement in world models. Review substrate: OpenReview. The submission site is live.\n\nTopics in scope\n\nWe welcome work that places continual adaptation at the center of world modeling, from architecture and memory to evaluation and embodied data acquisition.\n\n- Adaptive world model architectures\n- Test-time and lifelong learning\n- Memory for physical environments\n- Video prediction and generation\n- Spatial and temporal reasoning\n- Embodied data acquisition\n- Physical understanding\n- Evaluation protocols for continual improvement\n\nRecognition for contributed work: Selected papers will receive spotlights, with best paper and runner-up awards planned for the workshop.\n\nImportant dates:\n- Submission deadline: 29 August 2026, Anywhere on Earth\n- Author notification: 29 Sep 2026\n- Workshop: December 2026, Sydney, Australia", "cfp_status": "published", "topics": [ "Adaptive world model architectures", "Test-time and lifelong learning", "Memory for physical environments", "Video prediction and generation", "Spatial and temporal reasoning", "Embodied data acquisition", "Physical understanding", "Evaluation protocols for continual improvement" ], "important_dates": [ { "label": "Submission deadline", "date": "2026-08-29" }, { "label": "Author notification", "date": "2026-09-29" }, { "label": "Workshop", "date": "2026-12-10" } ], "organizers": [ "Laura Leal-Taixé (NVIDIA / University of Toronto)", "Xindi Wu (Princeton)", "Jonathan Lorraine (NVIDIA)", "Qianqian Wang (Harvard / Rhoda AI)", "Shengbang Tong (NYU / AMI Labs)", "Amir Bar (Meta / Imperial College London)" ], "speakers": [ "Danijar Hafner (Stealth Startup)", "Saining Xie (NYU / Google DeepMind)", "Yilun Du (Harvard)", "Homanga Bharadhwaj (Meta / JHU)", "Ming-Yu Liu (NVIDIA)", "Hao Ouyang (Ant Group LingBot)", "Yinghao Xu (Ant Group LingBot)" ], "host_url": "https://continual-world-models-workshop.github.io", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/CWM", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/CWM", "location": "Sydney, Australia", "city": "Sydney", "workshop_date": "2026-12-10", "contact": "xindiw@princeton.edu", "tracks": [ { "key": "CWM", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/CWM", "submission_dates_raw": "" } ], "group": "agents", "group_label": "Agents & Agentic Systems", "corpus": "NeurIPS 2026 Workshop on Continual World Models NeurIPS 2026 Workshop CWM A NeurIPS 2026 workshop bringing together computer vision, robotics, reinforcement learning, generative modeling, and cognitive science to define a research agenda for continual world models — systems that improve from the worlds they observe, imagine, and act within rather than remaining frozen predictors. Adaptive world model architectures Test-time and lifelong learning Memory for physical environments Video prediction and generation Spatial and temporal reasoning Embodied data acquisition Physical understanding Evaluation protocols for continual improvement Continual World Models\n\nA world model should change when the world does.\n\nThe missing continual axis\n\nWorld models increasingly encode rich motion and physical priors, yet most remain frozen predictors: trained once, then expected to generalize to whatever appears after deployment. Open-ended environments demand something more exacting, the capacity to notice prediction failures, absorb new evidence, and revise an evolving model of the world.\n\nThis workshop brings together computer vision, robotics, reinforcement learning, generative modeling, and cognitive science to define a research agenda for continual world models. We are interested in systems that do not merely extrapolate from fixed datasets, but improve from the worlds they observe, imagine, and act within.\n\nCall for papers\n\nWe invite 4-8 page non-archival submissions that advance our understanding of continual improvement in world models. Review substrate: OpenReview. The submission site is live.\n\nTopics in scope\n\nWe welcome work that places continual adaptation at the center of world modeling, from architecture and memory to evaluation and embodied data acquisition.\n\n- Adaptive world model architectures\n- Test-time and lifelong learning\n- Memory for physical environments\n- Video prediction and generation\n- Spatial and temporal reasoning\n- Embodied data acquisition\n- Physical understanding\n- Evaluation protocols for continual improvement\n\nRecognition for contributed work: Selected papers will receive spotlights, with best paper and runner-up awards planned for the workshop.\n\nImportant dates:\n- Submission deadline: 29 August 2026, Anywhere on Earth\n- Author notification: 29 Sep 2026\n- Workshop: December 2026, Sydney, Australia" }, { "key": "BiAlign", "title": "NeurIPS 2026 Workshop on Dynamic Alignment in Human-AI Coupled Systems", "subtitle": "BiAlign Workshop 2026", "summary": "The BiAlign @ NeurIPS 2026 workshop develops the theoretical, algorithmic, and evaluative foundations of dynamic alignment in human-AI coupled systems that learn, adapt, and co-evolve over time, examining safety at the level of the coupled human-AI system rather than the model alone.", "cfp_full": "This NeurIPS 2026 Workshop focuses on Dynamic Alignment in Human-AI Coupled Systems - advancing safety for human-AI systems that learn, adapt, and co-evolve over time. Large language models, multimodal models, and AI systems are increasingly shifting from single-turn response tools to long-term interactive actors in education, healthcare, scientific discovery, and other high-stakes domains. In these settings, AI systems do not merely affect task outcomes: they shape human trust, dependence, risk perception, values, and behavior, while human feedback and behavior in turn influence agents' reward signals, objectives, and policy updates. Real-world human-AI systems are therefore bidirectionally coupled and dynamically evolving. Building on the Bidirectional Human-AI Alignment framework (NeurIPS 2025 Position Paper) and our NeurIPS 2025 Tutorial, prior BiAlign Workshops at ICLR/CHI 2025, CHI 2026, and the Human-AI Alignment Course at NYU Shanghai, this workshop emphasizes two key shifts:\n- From AI-Level to Human-AI System-Level Safety: examining AI risks from the perspective of the broader human-AI coupled system, rather than focusing only on the AI model side;\n- From Static to Dynamic Alignment: accounting for continual learning, adaptation, and co-evolution within the coupled system, rather than treating alignment as a static, one-time objective.\n\nMainstream alignment methods often treat human feedback as exogenous, preferences as stable, and alignment as single-turn output improvement or preference matching. While useful for improving immediate model behavior, this paradigm does not fully capture long-term risks such as sycophancy, manipulation, overreliance, value-action gap, and delusional spirals, which arise from the mutual influence between human states and agent states. Consequently, the same AI system may lead to divergent outcomes depending on users' cognitive, emotional, and social contexts. AI alignment and safety should therefore be evaluated not only at the model level, but at the level of the coupled human-AI system, taking humans into account.\n\nThe core goals of this workshop are fivefold: (1) Establish Dynamic Alignment as a Core Research Problem for AI Safety and Learning; (2) Develop Foundations for Modeling, Measuring, and Controlling Alignment Dynamics; (3) Advance Evaluation Methods and Benchmarks beyond Single-Turn Settings; (4) Foster Interdisciplinary Exchange across Technical and Societal Fields; (5) Produce Concrete Community Outcomes and a Shared Research Agenda.\n\nThis workshop aims to establish the theoretical, algorithmic, and evaluative foundations of dynamic alignment in human-AI coupled systems that co-evolve over time.\n\nCall for Papers\n\nWe invite researchers and practitioners from academia and industry to join our Workshop on Dynamic Alignment in Human-AI Coupled Systems at NeurIPS 2026. As AI systems shift from single-turn response tools to long-term interactive actors in high-stakes domains, alignment can no longer be treated as a static, one-time objective: human states, agent policies, feedback signals, objectives, and evaluation criteria co-evolve over time. This workshop provides a forum to develop the theoretical, algorithmic, and evaluative foundations of dynamic alignment - spanning AI alignment, machine learning, human-AI interaction, cognitive science, social computing, law, governance, and philosophy. The one-day hybrid workshop features five keynotes, a panel discussion, spotlight talks, two poster sessions, and Structured Thematic Breakout Discussions at on-site interaction stations, whose outcomes will feed into a potential post-workshop white paper. We welcome submissions from all relevant disciplines; accepted work will be presented as spotlight talks or posters, as decided by the program committee. Key workshop topics include:\n\n- Foundations of Dynamic Alignment: Conceptual and theoretical work clarifying alignment in systems where human states, agent policies, feedback signals, objectives, and environments co-evolve. Research Questions: How should alignment be defined and formalized when humans are not fixed feedback oracles and agents are not static tools? What theoretical tools characterize stability, path dependence, and long-run outcomes of coupled human-AI dynamics? Keyword Examples: stability, path dependence, endogenous feedback, influenceable preferences, dynamical systems, etc.\n- Specification, Objectives, and Evolving Preferences: Methods for specifying and updating goals, values, norms, preferences, and safety constraints under changing human and social contexts. Research Questions: How can objectives and safety constraints remain well-specified as human and social contexts change? How should systems handle disagreement, preference drift, collective values, and normative uncertainty? Keyword Examples: preference drift, value specification, normative uncertainty, collective values, pluralistic alignment, etc.\n- Learning, Adaptation, and Feedback Dynamics: Studies of how agents learn from repeated human interaction through RL, human-in-the-loop learning, memory, and lifelong adaptation. Research Questions: How do agents learn and adapt under non-stationary, noisy, or strategic human feedback? What learning dynamics emerge from repeated interaction, memory, and lifelong adaptation? Keyword Examples: RLHF, human-in-the-loop learning, non-stationary feedback, lifelong learning, multi-turn RL, etc.\n- Evaluation, Benchmarks, and Failure Discovery: Metrics, benchmarks, simulations, audits, and red-teaming methods for long-horizon and interactive alignment. Research Questions: How can we detect and measure safety-critical failures that only emerge over long-horizon interaction, such as sycophancy, overreliance, trust miscalibration, value drift, and manipulation? What benchmarks and simulations capture coupled human-AI dynamics? Keyword Examples: long-horizon evaluation, red-teaming, audits, dark patterns, sycophancy, overreliance, multi-agent risks, etc.\n- Control, Intervention, and Monitoring: Approaches for maintaining or restoring alignment after deployment. Research Questions: How can alignment be maintained or restored once systems are deployed and co-evolving with users? What runtime monitoring, oversight, and rollback mechanisms scale to real-world coupled systems? Keyword Examples: runtime monitoring, scalable oversight, adaptive safeguards, rollback, human control interfaces, incident analysis, etc.\n- Collective, Multi-Agent, and Societal Dynamics: Work on alignment across multiple humans, agents, organizations, or institutions. Research Questions: How does alignment behave across populations of humans and agents - under coordination, competition, collusion, and social influence? How do performative effects, governance interfaces, and institutional constraints shape alignment at scale? Keyword Examples: multi-agent systems, social influence, performative prediction, governance, institutional constraints, etc.\n- Alignment in Domain-Driven and High-Stakes Applications: Research grounded in real-world settings where long-term interaction and safety constraints shape alignment. Research Questions: How do long-term interaction and safety constraints shape alignment in domains such as education, healthcare, mental health, scientific discovery, robotics, recommender systems, and public-sector decision-making? Keyword Examples: healthcare, mental health, education, agentic science, robotics, recommender systems, public-sector AI, etc.\n\nSubmission Format: We call for 2-page (tiny), 4-page (short), and 9-page (long) papers, excluding references, fully anonymized. Non-archival. Accepted work will be presented as spotlight talks (4 selected) or posters across two poster sessions. All accepted papers will be published on the workshop website (non-archival). The program committee will select 3-5 Workshop Awards announced before closing remarks. Each submission receives three reviews; conflicts of interest are managed per NeurIPS Workshop requirements. We plan to invite participants to collaborate on a joint post-workshop white paper on dynamic alignment.\n\nImportant Dates\n- Submission: August 29, 2026\n- Notification: September 29, 2026\n- Camera ready: October 29, 2026\n- NeurIPS Workshop: December 11 or 12, 2026", "cfp_status": "published", "topics": [ "Foundations of Dynamic Alignment", "Specification, Objectives, and Evolving Preferences", "Learning, Adaptation, and Feedback Dynamics", "Evaluation, Benchmarks, and Failure Discovery", "Control, Intervention, and Monitoring", "Collective, Multi-Agent, and Societal Dynamics", "Alignment in Domain-Driven and High-Stakes Applications" ], "important_dates": [ { "label": "Submission", "date": "2026-08-29" }, { "label": "Notification", "date": "2026-09-29" }, { "label": "Camera ready", "date": "2026-10-29" }, { "label": "Workshop", "date": "December 11 or 12, 2026" } ], "organizers": [ "Hua Shen (NYU Shanghai / New York University)", "Divy Thakkar (Google DeepMind)", "Gordon Dai (NYU Shanghai / New York University)", "Vivek Myers (UC Berkeley)", "Nick Haber (Stanford University)", "Joan Bruna (New York University)", "Dawn Song (UC Berkeley)", "Yoshua Bengio (Mila / LawZero / Universite de Montreal)" ], "speakers": [], "host_url": "https://bialign-workshop.github.io/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/BiAlign", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/BiAlign", "location": "Sydney, Australia", "city": "Sydney", "workshop_date": "2026-12-10", "contact": "hs3645@nyu.edu", "tracks": [ { "key": "BiAlign", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/BiAlign", "submission_dates_raw": "" } ], "group": "trust", "group_label": "Trustworthy & Safe AI", "corpus": "NeurIPS 2026 Workshop on Dynamic Alignment in Human-AI Coupled Systems BiAlign Workshop 2026 The BiAlign @ NeurIPS 2026 workshop develops the theoretical, algorithmic, and evaluative foundations of dynamic alignment in human-AI coupled systems that learn, adapt, and co-evolve over time, examining safety at the level of the coupled human-AI system rather than the model alone. Foundations of Dynamic Alignment Specification, Objectives, and Evolving Preferences Learning, Adaptation, and Feedback Dynamics Evaluation, Benchmarks, and Failure Discovery Control, Intervention, and Monitoring Collective, Multi-Agent, and Societal Dynamics Alignment in Domain-Driven and High-Stakes Applications This NeurIPS 2026 Workshop focuses on Dynamic Alignment in Human-AI Coupled Systems - advancing safety for human-AI systems that learn, adapt, and co-evolve over time. Large language models, multimodal models, and AI systems are increasingly shifting from single-turn response tools to long-term interactive actors in education, healthcare, scientific discovery, and other high-stakes domains. In these settings, AI systems do not merely affect task outcomes: they shape human trust, dependence, risk perception, values, and behavior, while human feedback and behavior in turn influence agents' reward signals, objectives, and policy updates. Real-world human-AI systems are therefore bidirectionally coupled and dynamically evolving. Building on the Bidirectional Human-AI Alignment framework (NeurIPS 2025 Position Paper) and our NeurIPS 2025 Tutorial, prior BiAlign Workshops at ICLR/CHI 2025, CHI 2026, and the Human-AI Alignment Course at NYU Shanghai, this workshop emphasizes two key shifts:\n- From AI-Level to Human-AI System-Level Safety: examining AI risks from the perspective of the broader human-AI coupled system, rather than focusing only on the AI model side;\n- From Static to Dynamic Alignment: accounting for continual learning, adaptation, and co-evolution within the coupled system, rather than treating alignment as a static, one-time objective.\n\nMainstream alignment methods often treat human feedback as exogenous, preferences as stable, and alignment as single-turn output improvement or preference matching. While useful for improving immediate model behavior, this paradigm does not fully capture long-term risks such as sycophancy, manipulation, overreliance, value-action gap, and delusional spirals, which arise from the mutual influence between human states and agent states. Consequently, the same AI system may lead to divergent outcomes depending on users' cognitive, emotional, and social contexts. AI alignment and safety should therefore be evaluated not only at the model level, but at the level of the coupled human-AI system, taking humans into account.\n\nThe core goals of this workshop are fivefold: (1) Establish Dynamic Alignment as a Core Research Problem for AI Safety and Learning; (2) Develop Foundations for Modeling, Measuring, and Controlling Alignment Dynamics; (3) Advance Evaluation Methods and Benchmarks beyond Single-Turn Settings; (4) Foster Interdisciplinary Exchange across Technical and Societal Fields; (5) Produce Concrete Community Outcomes and a Shared Research Agenda.\n\nThis workshop aims to establish the theoretical, algorithmic, and evaluative foundations of dynamic alignment in human-AI coupled systems that co-evolve over time.\n\nCall for Papers\n\nWe invite researchers and practitioners from academia and industry to join our Workshop on Dynamic Alignment in Human-AI Coupled Systems at NeurIPS 2026. As AI systems shift from single-turn response tools to long-term interactive actors in high-stakes domains, alignment can no longer be treated as a static, one-time objective: human states, agent policies, feedback signals, objectives, and evaluation criteria co-evolve over time. This workshop provides a forum to develop the theoretical, algorithmic, and evaluative foundations of dynamic alignment - spanning AI alignment, machine learning, human-AI interaction, cognitive science, social computing, law, governance, and philosophy. The one-day hybrid workshop features five keynotes, a panel discussion, spotlight talks, two poster sessions, and Structured Thematic Breakout Discussions at on-site interaction stations, whose outcomes will feed into a potential post-workshop white paper. We welcome submissions from all relevant disciplines; accepted work will be presented as spotlight talks or posters, as decided by the program committee. Key workshop topics include:\n\n- Foundations of Dynamic Alignment: Conceptual and theoretical work clarifying alignment in systems where human states, agent policies, feedback signals, objectives, and environments co-evolve. Research Questions: How should alignment be defined and formalized when humans are not fixed feedback oracles and agents are not static tools? What theoretical tools characterize stability, path dependence, and long-run outcomes of coupled human-AI dynamics? Keyword Examples: stability, path dependence, endogenous feedback, influenceable preferences, dynamical systems, etc.\n- Specification, Objectives, and Evolving Preferences: Methods for specifying and updating goals, values, norms, preferences, and safety constraints under changing human and social contexts. Research Questions: How can objectives and safety constraints remain well-specified as human and social contexts change? How should systems handle disagreement, preference drift, collective values, and normative uncertainty? Keyword Examples: preference drift, value specification, normative uncertainty, collective values, pluralistic alignment, etc.\n- Learning, Adaptation, and Feedback Dynamics: Studies of how agents learn from repeated human interaction through RL, human-in-the-loop learning, memory, and lifelong adaptation. Research Questions: How do agents learn and adapt under non-stationary, noisy, or strategic human feedback? What learning dynamics emerge from repeated interaction, memory, and lifelong adaptation? Keyword Examples: RLHF, human-in-the-loop learning, non-stationary feedback, lifelong learning, multi-turn RL, etc.\n- Evaluation, Benchmarks, and Failure Discovery: Metrics, benchmarks, simulations, audits, and red-teaming methods for long-horizon and interactive alignment. Research Questions: How can we detect and measure safety-critical failures that only emerge over long-horizon interaction, such as sycophancy, overreliance, trust miscalibration, value drift, and manipulation? What benchmarks and simulations capture coupled human-AI dynamics? Keyword Examples: long-horizon evaluation, red-teaming, audits, dark patterns, sycophancy, overreliance, multi-agent risks, etc.\n- Control, Intervention, and Monitoring: Approaches for maintaining or restoring alignment after deployment. Research Questions: How can alignment be maintained or restored once systems are deployed and co-evolving with users? What runtime monitoring, oversight, and rollback mechanisms scale to real-world coupled systems? Keyword Examples: runtime monitoring, scalable oversight, adaptive safeguards, rollback, human control interfaces, incident analysis, etc.\n- Collective, Multi-Agent, and Societal Dynamics: Work on alignment across multiple humans, agents, organizations, or institutions. Research Questions: How does alignment behave across populations of humans and agents - under coordination, competition, collusion, and social influence? How do performative effects, governance interfaces, and institutional constraints shape alignment at scale? Keyword Examples: multi-agent systems, social influence, performative prediction, governance, institutional constraints, etc.\n- Alignment in Domain-Driven and High-Stakes Applications: Research grounded in real-world settings where long-term interaction and safety constraints shape alignment. Research Questions: How do long-term interaction and safety constraints shape alignment in domains such as education, healthcare, mental health, scientific discovery, robotics, recommender systems, and public-sector decision-making? Keyword Examples: healthcare, mental health, education, agentic science, robotics, recommender systems, public-sector AI, etc.\n\nSubmission Format: We call for 2-page (tiny), 4-page (short), and 9-page (long) papers, excluding references, fully anonymized. Non-archival. Accepted work will be presented as spotlight talks (4 selected) or posters across two poster sessions. All accepted papers will be published on the workshop website (non-archival). The program committee will select 3-5 Workshop Awards announced before closing remarks. Each submission receives three reviews; conflicts of interest are managed per NeurIPS Workshop requirements. We plan to invite participants to collaborate on a joint post-workshop white paper on dynamic alignment.\n\nImportant Dates\n- Submission: August 29, 2026\n- Notification: September 29, 2026\n- Camera ready: October 29, 2026\n- NeurIPS Workshop: December 11 or 12, 2026" }, { "key": "E-values", "title": "NeurIPS 2026 Workshop on E-values: From Statistics To ML", "subtitle": "NeurIPS 2026 Workshop E-values", "summary": "A workshop bringing together statistics, machine learning, game-theoretic probability, and sequential analysis to advance the use of e-values for anytime-valid inference, continuous monitoring, and data-adaptive decision making in modern ML systems.", "cfp_full": "E-Values: From Statistics to ML\nBringing together statistics, machine learning, game-theoretic probability, and sequential analysis to advance e-values in modern ML systems.\n\nRecent years have seen a paradigm shift in hypothesis testing and uncertainty quantification through e-values: non-negative random variables whose null expectation is bounded by one. They support anytime-valid inference, continuous monitoring, and data-adaptive decision making, making them especially relevant for modern ML systems that are evaluated, audited, and updated sequentially. This workshop will bring these threads together through invited talks, contributed talks, posters, and a panel discussion.\n\nTopics\nFrom safe testing to deployed ML systems\nWe welcome polished results as well as early ideas, emerging connections, and practical case studies that help define the role of e-values in machine learning.\n\nFoundations & Theory\n- E-values, e-processes, test martingales, and game-theoretic probability\n- E-values as evidence, including links to p-values and Bayes factors\n- E-values, Bayesian methods, pseudo-Bayesian methods, and e-posteriors\n- Multiple testing, FDR, FWER, and empirical-Bayes connections\n- Compositional e-value guarantees across multi-stage or multi-agent workflows\n\nSequential & Adaptive Inference\n- Sequential monitoring, evidence aggregation, and stopping rules with e-values\n- Anytime-valid confidence intervals, prediction sets, and safe testing\n- Optional stopping, optional continuation, and continuous monitoring\n- Conformal prediction, adaptive coverage, and e-based prediction sets\n- A/B testing, adaptive experimentation, bandits, and online learning regret\n\nML Applications, Auditing & Case Studies\n- Uncertainty under prompt variation, adaptivity, and distribution shift\n- E-values for LLM evaluation, foundation models, and deployed AI systems\n- Auditing and verification for tool use, external feedback loops, fairness, and privacy\n- Applications in science, medicine, safety-critical systems, and decision support\n- Software, computation, and practical case studies with e-values\n\nCall for Papers\nWe invite submissions on e-values and modern inference for ML.\nSubmissions may cover theoretical, algorithmic, or practical aspects of e-values and related ideas. The workshop is non-archival.\n\nSubmission Guidelines\n- Short papers up to 4 pages, excluding references and optional appendices.\n- Submit your papers using the NeurIPS 2026 LaTeX template (with the dblblindworkshop option).\n- Accepted papers will be presented as posters, with a subset selected for oral talks.\n\nSubmission Portal\nSubmit papers through the workshop OpenReview page. Please create your OpenReview profile at least two weeks before the submission deadline.\n\nImportant Dates\nAll deadlines are 23:59 Anywhere on Earth (AOE).\n- Submission deadline: August 29, 2026\n- Notification of acceptance: September 29, 2026\n- Camera-ready due: TBD\n- Workshop date: December 12 or 13, 2026 (TBD)", "cfp_status": "published", "topics": [ "E-values, e-processes, test martingales, and game-theoretic probability", "E-values as evidence, including links to p-values and Bayes factors", "E-values, Bayesian methods, pseudo-Bayesian methods, and e-posteriors", "Multiple testing, FDR, FWER, and empirical-Bayes connections", "Compositional e-value guarantees across multi-stage or multi-agent workflows", "Sequential monitoring, evidence aggregation, and stopping rules with e-values", "Anytime-valid confidence intervals, prediction sets, and safe testing", "Optional stopping, optional continuation, and continuous monitoring", "Conformal prediction, adaptive coverage, and e-based prediction sets", "A/B testing, adaptive experimentation, bandits, and online learning regret", "Uncertainty under prompt variation, adaptivity, and distribution shift", "E-values for LLM evaluation, foundation models, and deployed AI systems", "Auditing and verification for tool use, external feedback loops, fairness, and privacy", "Applications in science, medicine, safety-critical systems, and decision support", "Software, computation, and practical case studies with e-values" ], "important_dates": [ { "label": "Submission deadline", "date": "2026-08-29" }, { "label": "Notification of acceptance", "date": "2026-09-29" }, { "label": "Camera-ready due", "date": "TBD" }, { "label": "Workshop date", "date": "December 12 or 13, 2026 (TBD)" } ], "organizers": [ "Shubhada Agrawal (Indian Institute of Science, Bangalore)", "Sebastian Arnold (CWI, Amsterdam)", "Yo Joong (YJ) Choe (INSEAD, Singapore)", "Peter Grünwald (CWI, Amsterdam and Leiden University)", "Aaditya Ramdas (Stanford University)" ], "speakers": [ "Michael I. Jordan (UC Berkeley and INRIA Paris)", "Rianne de Heide (University of Twente and CWI)", "Eugenio Clerico (University of Oxford)", "Emilie Kaufmann (CNRS, Université de Lille)", "Nick Koning (Erasmus University Rotterdam)" ], "host_url": "https://e-values-workshop.github.io/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/E-values", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/E-values", "location": "Paris, France", "city": "Paris", "workshop_date": "2026-12-12", "contact": "evalue.workshop.2026@gmail.com", "tracks": [ { "key": "E-values", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/E-values", "submission_dates_raw": "" } ], "group": "theory", "group_label": "Theory & Methods", "corpus": "NeurIPS 2026 Workshop on E-values: From Statistics To ML NeurIPS 2026 Workshop E-values A workshop bringing together statistics, machine learning, game-theoretic probability, and sequential analysis to advance the use of e-values for anytime-valid inference, continuous monitoring, and data-adaptive decision making in modern ML systems. E-values, e-processes, test martingales, and game-theoretic probability E-values as evidence, including links to p-values and Bayes factors E-values, Bayesian methods, pseudo-Bayesian methods, and e-posteriors Multiple testing, FDR, FWER, and empirical-Bayes connections Compositional e-value guarantees across multi-stage or multi-agent workflows Sequential monitoring, evidence aggregation, and stopping rules with e-values Anytime-valid confidence intervals, prediction sets, and safe testing Optional stopping, optional continuation, and continuous monitoring Conformal prediction, adaptive coverage, and e-based prediction sets A/B testing, adaptive experimentation, bandits, and online learning regret Uncertainty under prompt variation, adaptivity, and distribution shift E-values for LLM evaluation, foundation models, and deployed AI systems Auditing and verification for tool use, external feedback loops, fairness, and privacy Applications in science, medicine, safety-critical systems, and decision support Software, computation, and practical case studies with e-values E-Values: From Statistics to ML\nBringing together statistics, machine learning, game-theoretic probability, and sequential analysis to advance e-values in modern ML systems.\n\nRecent years have seen a paradigm shift in hypothesis testing and uncertainty quantification through e-values: non-negative random variables whose null expectation is bounded by one. They support anytime-valid inference, continuous monitoring, and data-adaptive decision making, making them especially relevant for modern ML systems that are evaluated, audited, and updated sequentially. This workshop will bring these threads together through invited talks, contributed talks, posters, and a panel discussion.\n\nTopics\nFrom safe testing to deployed ML systems\nWe welcome polished results as well as early ideas, emerging connections, and practical case studies that help define the role of e-values in machine learning.\n\nFoundations & Theory\n- E-values, e-processes, test martingales, and game-theoretic probability\n- E-values as evidence, including links to p-values and Bayes factors\n- E-values, Bayesian methods, pseudo-Bayesian methods, and e-posteriors\n- Multiple testing, FDR, FWER, and empirical-Bayes connections\n- Compositional e-value guarantees across multi-stage or multi-agent workflows\n\nSequential & Adaptive Inference\n- Sequential monitoring, evidence aggregation, and stopping rules with e-values\n- Anytime-valid confidence intervals, prediction sets, and safe testing\n- Optional stopping, optional continuation, and continuous monitoring\n- Conformal prediction, adaptive coverage, and e-based prediction sets\n- A/B testing, adaptive experimentation, bandits, and online learning regret\n\nML Applications, Auditing & Case Studies\n- Uncertainty under prompt variation, adaptivity, and distribution shift\n- E-values for LLM evaluation, foundation models, and deployed AI systems\n- Auditing and verification for tool use, external feedback loops, fairness, and privacy\n- Applications in science, medicine, safety-critical systems, and decision support\n- Software, computation, and practical case studies with e-values\n\nCall for Papers\nWe invite submissions on e-values and modern inference for ML.\nSubmissions may cover theoretical, algorithmic, or practical aspects of e-values and related ideas. The workshop is non-archival.\n\nSubmission Guidelines\n- Short papers up to 4 pages, excluding references and optional appendices.\n- Submit your papers using the NeurIPS 2026 LaTeX template (with the dblblindworkshop option).\n- Accepted papers will be presented as posters, with a subset selected for oral talks.\n\nSubmission Portal\nSubmit papers through the workshop OpenReview page. Please create your OpenReview profile at least two weeks before the submission deadline.\n\nImportant Dates\nAll deadlines are 23:59 Anywhere on Earth (AOE).\n- Submission deadline: August 29, 2026\n- Notification of acceptance: September 29, 2026\n- Camera-ready due: TBD\n- Workshop date: December 12 or 13, 2026 (TBD)" }, { "key": "FLLMPT", "title": "NeurIPS 2026 Workshop on Foundations of LLM Post-Training in Changing Environments", "subtitle": "FLLMPT NeurIPS 2026", "summary": "A workshop developing principled theoretical foundations for large language model post-training under task evolution and non-stationarity, bringing together researchers from machine learning theory, reinforcement learning, and AI safety.", "cfp_full": "About the Workshop\nLarge language models (LLMs) are routinely adapted to downstream applications through post-training methods, such as instruction tuning and domain adaptation. Yet in real-world deployment, downstream tasks rarely remain fixed: objectives shift, data distributions drift, feedback signals evolve, and evaluation standards change over time. Post-training therefore becomes a process of repeated adaptation in non-stationary environments.\n\nDespite its central role in modern foundation models, the theoretical foundations of this adaptive post-training paradigm remain limited. Current practices are largely heuristic, with incomplete understanding of statistical identifiability, optimization dynamics, robustness to misspecification, and trade-offs between adaptation and capability preservation. These gaps are particularly consequential in safety-critical settings, where unintended regressions or feedback loops may arise under evolving conditions.\n\nThis workshop will develop principled foundations for LLM post-training under task evolution. It will bring together researchers from machine learning theory, reinforcement learning, and AI safety to develop principled foundations for this.\n\nMotivation and Timeliness\nLLMs are inherently multitask systems, capable of performing a wide range of tasks using a single set of parameters. They undergo post-training adaptation through methods like fine-tuning and preference-based updates. However, downstream tasks evolve post-deployment, creating non-stationarity that demands continual updates. The workshop addresses a critical gap: limited theoretical understanding of when adaptation succeeds, when it fails, or how it affects previously learned capabilities. This matters especially for safety-critical applications, as the International AI Safety Report identifies theoretical understanding as essential for reasoning about robustness under task evolution.\n\nWorkshop Goals and Topics\nResearchers should submit work on LLM post-training in evolving environments, including:\n- Preference feedback as data: Statistical modeling of preference signals, identifiability, heterogeneous annotators, and noise treatment.\n- Robustness and valid inference: Conditions ensuring post-training robustness, uncertainty quantification, calibration, and principled evaluation.\n- Adaptive data collection and feedback loops: Sequential collection effects, selection bias, active query design, and evolving evaluation criteria.\n- Adaptation mechanisms and limits: Theoretical understanding of parameter-efficient adaptation, modular updates, selective fine-tuning, capability preservation, and failure modes.\n\nSubmission Instructions\nFormat: PDF using NeurIPS 2026 style. Long papers: 8 pages maximum (excluding references/appendices). Short papers: 4 pages maximum. Unlimited references and supplementary material.\nReview Process: All submissions will undergo double-blind peer review. Reviewers assess novelty, significance, technical quality, and clarity.\nDual-submission Policy: FLLMPT 2026 is a non-archival workshop. We welcome submissions of work currently under review at other venues and work that has previously been published, provided it has been updated or recontextualised for the workshop audience.\nPortal: Submit via OpenReview.\n\nKey Dates\n- Submission Portal Opens: Open (OpenReview)\n- Abstract Registration: 2026/08/23 11:59 AoE (One week before paper deadline)\n- Paper Submission: 2026/08/30 11:59 AoE (Firm deadline)\n- Review Period: Sept–Oct 2026 (Double-blind)\n- Author Notification: Oct 2026 (TBA)\n- Camera-Ready: Nov 2026 (TBA) (Non-archival)\n- Workshop Date: Dec 2026 (Paris, France)", "cfp_status": "published", "topics": [ "Preference feedback as data: statistical modeling of preference signals, identifiability, heterogeneous annotators, and noise treatment", "Robustness and valid inference: conditions ensuring post-training robustness, uncertainty quantification, calibration, and principled evaluation", "Adaptive data collection and feedback loops: sequential collection effects, selection bias, active query design, and evolving evaluation criteria", "Adaptation mechanisms and limits: parameter-efficient adaptation, modular updates, selective fine-tuning, capability preservation, and failure modes" ], "important_dates": [ { "label": "Abstract Registration", "date": "2026-08-23" }, { "label": "Paper Submission", "date": "2026-08-30" }, { "label": "Author Notification", "date": "Oct 2026 (TBA)" }, { "label": "Camera-Ready", "date": "Nov 2026 (TBA)" }, { "label": "Workshop Date", "date": "Dec 2026" } ], "organizers": [], "speakers": [], "host_url": "https://www.fllmpt-work.shop/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/FLLMPT", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/FLLMPT", "location": "NeurIPS Paris 2026", "city": "Paris", "workshop_date": "2026-12-08", "contact": "contact@fllmpt-work.shop", "tracks": [ { "key": "FLLMPT", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/FLLMPT", "submission_dates_raw": "" } ], "group": "fm", "group_label": "Foundation Models & LLMs", "corpus": "NeurIPS 2026 Workshop on Foundations of LLM Post-Training in Changing Environments FLLMPT NeurIPS 2026 A workshop developing principled theoretical foundations for large language model post-training under task evolution and non-stationarity, bringing together researchers from machine learning theory, reinforcement learning, and AI safety. Preference feedback as data: statistical modeling of preference signals, identifiability, heterogeneous annotators, and noise treatment Robustness and valid inference: conditions ensuring post-training robustness, uncertainty quantification, calibration, and principled evaluation Adaptive data collection and feedback loops: sequential collection effects, selection bias, active query design, and evolving evaluation criteria Adaptation mechanisms and limits: parameter-efficient adaptation, modular updates, selective fine-tuning, capability preservation, and failure modes About the Workshop\nLarge language models (LLMs) are routinely adapted to downstream applications through post-training methods, such as instruction tuning and domain adaptation. Yet in real-world deployment, downstream tasks rarely remain fixed: objectives shift, data distributions drift, feedback signals evolve, and evaluation standards change over time. Post-training therefore becomes a process of repeated adaptation in non-stationary environments.\n\nDespite its central role in modern foundation models, the theoretical foundations of this adaptive post-training paradigm remain limited. Current practices are largely heuristic, with incomplete understanding of statistical identifiability, optimization dynamics, robustness to misspecification, and trade-offs between adaptation and capability preservation. These gaps are particularly consequential in safety-critical settings, where unintended regressions or feedback loops may arise under evolving conditions.\n\nThis workshop will develop principled foundations for LLM post-training under task evolution. It will bring together researchers from machine learning theory, reinforcement learning, and AI safety to develop principled foundations for this.\n\nMotivation and Timeliness\nLLMs are inherently multitask systems, capable of performing a wide range of tasks using a single set of parameters. They undergo post-training adaptation through methods like fine-tuning and preference-based updates. However, downstream tasks evolve post-deployment, creating non-stationarity that demands continual updates. The workshop addresses a critical gap: limited theoretical understanding of when adaptation succeeds, when it fails, or how it affects previously learned capabilities. This matters especially for safety-critical applications, as the International AI Safety Report identifies theoretical understanding as essential for reasoning about robustness under task evolution.\n\nWorkshop Goals and Topics\nResearchers should submit work on LLM post-training in evolving environments, including:\n- Preference feedback as data: Statistical modeling of preference signals, identifiability, heterogeneous annotators, and noise treatment.\n- Robustness and valid inference: Conditions ensuring post-training robustness, uncertainty quantification, calibration, and principled evaluation.\n- Adaptive data collection and feedback loops: Sequential collection effects, selection bias, active query design, and evolving evaluation criteria.\n- Adaptation mechanisms and limits: Theoretical understanding of parameter-efficient adaptation, modular updates, selective fine-tuning, capability preservation, and failure modes.\n\nSubmission Instructions\nFormat: PDF using NeurIPS 2026 style. Long papers: 8 pages maximum (excluding references/appendices). Short papers: 4 pages maximum. Unlimited references and supplementary material.\nReview Process: All submissions will undergo double-blind peer review. Reviewers assess novelty, significance, technical quality, and clarity.\nDual-submission Policy: FLLMPT 2026 is a non-archival workshop. We welcome submissions of work currently under review at other venues and work that has previously been published, provided it has been updated or recontextualised for the workshop audience.\nPortal: Submit via OpenReview.\n\nKey Dates\n- Submission Portal Opens: Open (OpenReview)\n- Abstract Registration: 2026/08/23 11:59 AoE (One week before paper deadline)\n- Paper Submission: 2026/08/30 11:59 AoE (Firm deadline)\n- Review Period: Sept–Oct 2026 (Double-blind)\n- Author Notification: Oct 2026 (TBA)\n- Camera-Ready: Nov 2026 (TBA) (Non-archival)\n- Workshop Date: Dec 2026 (Paris, France)" }, { "key": "Interp4Discovery", "title": "NeurIPS 2026 Workshop on Interpretability for Discovery", "subtitle": "Interp4Discovery", "summary": "A NeurIPS 2026 workshop reframing model interpretability as a tool for scientific discovery, bringing interpretability researchers and domain scientists together to turn what models encode into novel, testable knowledge that experts can validate.", "cfp_full": "What do AI models know that we don't? We explore how model interpretability can turn learned representations into novel, testable knowledge.\n\nAbout the Workshop\nAI models now match or exceed human experts across domains from protein structure prediction and clinical forecasting to astronomy, climate science, animal communication, and strategic games. These models appear to have learned patterns humans haven't yet articulated. This workshop reframes interpretability as a tool for discovery, bringing interpretability researchers and domain scientists together to turn what models encode into knowledge experts can test and validate. Interpretability is more than a debugging tool. It can bridge black-box models and scientific breakthroughs across unfamiliar domains, architectures, and modalities. From prediction to discovery: explanation is only the beginning.\n\nCore Questions:\n- Adapting modalities — Methods for interpreting architectures and data types beyond standard language and vision approaches\n- Translating knowledge — Converting interpretable representations into actionable, testable insights\n- Open scientific problems — Questions interpretability can address that remain unknown to humans\n\nResearch Scope Topics:\n- Methods and models for knowledge discovery across unfamiliar systems\n- Interpretability-driven discovery through empirical case studies\n- Perspectives on novel knowledge including epistemology and validation\n- Failure case analyses clarifying interpretability's limits\n\nCall for Papers\nThe workshop invites original methodological, empirical, theoretical, and position work on using interpretability to uncover new knowledge. Failure cases and negative results are welcome.\n\nPaper Format\nPapers must be up to 5 pages of main text (an additional page is allowed for camera-ready versions). References and appendices don't count toward the page limit, but the main text must be self-contained. Review is double-blind and the workshop is non-archival. Submit through OpenReview as a single PDF written in English, using the NeurIPS 2026 workshop LaTeX template. Remove all author-identifying information.\n\nReproducibility: Authors are strongly encouraged to open source relevant code, models, prompts, data, and interactive demos.\n\nEligibility: Non-archival venue work is welcome. Work already accepted at an archival venue is excluded, except NeurIPS 2026 papers through fast-track consideration.\n\nReview Policy: Double-blind review process. Every submission requires a short statement covering potential societal impacts and suggested mitigations.\n\nPresentation: Accepted papers appear as posters, with some selected for oral or spotlight presentations.\n\nCurrent submission requirements are tentative, and additional formats may be added. All deadlines follow Anywhere on Earth (AoE).\n\nImportant Dates:\n- Submission deadline: August 29, 2026, 11:59 PM AoE\n- Author notification: September 29, 2026\n- Workshop event: December 12 or 13, 2026, Atlanta, Georgia, USA", "cfp_status": "published", "topics": [ "Methods and models for knowledge discovery across unfamiliar systems", "Interpretability-driven discovery through empirical case studies", "Perspectives on novel knowledge including epistemology and validation", "Failure case analyses clarifying interpretability's limits", "Interpreting architectures and data types beyond standard language and vision", "Translating interpretable representations into actionable, testable insights", "Open scientific problems that interpretability can address" ], "important_dates": [ { "label": "Submission Deadline", "date": "2026-08-29" }, { "label": "Author Notification", "date": "2026-09-29" }, { "label": "Workshop", "date": "2026-12-12" } ], "organizers": [ "Xiaoyan Bai (University of Chicago)", "Yonatan Belinkov (Technion)", "Ekdeep Singh Lubana (Goodfire AI)", "Yaniv Nikankin (Technion)", "Chenhao Tan (University of Chicago)", "Amirtha Varshini A S (Montai Therapeutics)" ], "speakers": [], "host_url": "https://interpretability4discovery.github.io/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/Interp4Discovery", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/Interp4Discovery", "location": "Atlanta, Georgia, USA", "city": "Atlanta", "workshop_date": "2026-12-12", "contact": "interp4discovery@gmail.com", "tracks": [ { "key": "Interp4Discovery", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/Interp4Discovery", "submission_dates_raw": "" } ], "group": "trust", "group_label": "Trustworthy & Safe AI", "corpus": "NeurIPS 2026 Workshop on Interpretability for Discovery Interp4Discovery A NeurIPS 2026 workshop reframing model interpretability as a tool for scientific discovery, bringing interpretability researchers and domain scientists together to turn what models encode into novel, testable knowledge that experts can validate. Methods and models for knowledge discovery across unfamiliar systems Interpretability-driven discovery through empirical case studies Perspectives on novel knowledge including epistemology and validation Failure case analyses clarifying interpretability's limits Interpreting architectures and data types beyond standard language and vision Translating interpretable representations into actionable, testable insights Open scientific problems that interpretability can address What do AI models know that we don't? We explore how model interpretability can turn learned representations into novel, testable knowledge.\n\nAbout the Workshop\nAI models now match or exceed human experts across domains from protein structure prediction and clinical forecasting to astronomy, climate science, animal communication, and strategic games. These models appear to have learned patterns humans haven't yet articulated. This workshop reframes interpretability as a tool for discovery, bringing interpretability researchers and domain scientists together to turn what models encode into knowledge experts can test and validate. Interpretability is more than a debugging tool. It can bridge black-box models and scientific breakthroughs across unfamiliar domains, architectures, and modalities. From prediction to discovery: explanation is only the beginning.\n\nCore Questions:\n- Adapting modalities — Methods for interpreting architectures and data types beyond standard language and vision approaches\n- Translating knowledge — Converting interpretable representations into actionable, testable insights\n- Open scientific problems — Questions interpretability can address that remain unknown to humans\n\nResearch Scope Topics:\n- Methods and models for knowledge discovery across unfamiliar systems\n- Interpretability-driven discovery through empirical case studies\n- Perspectives on novel knowledge including epistemology and validation\n- Failure case analyses clarifying interpretability's limits\n\nCall for Papers\nThe workshop invites original methodological, empirical, theoretical, and position work on using interpretability to uncover new knowledge. Failure cases and negative results are welcome.\n\nPaper Format\nPapers must be up to 5 pages of main text (an additional page is allowed for camera-ready versions). References and appendices don't count toward the page limit, but the main text must be self-contained. Review is double-blind and the workshop is non-archival. Submit through OpenReview as a single PDF written in English, using the NeurIPS 2026 workshop LaTeX template. Remove all author-identifying information.\n\nReproducibility: Authors are strongly encouraged to open source relevant code, models, prompts, data, and interactive demos.\n\nEligibility: Non-archival venue work is welcome. Work already accepted at an archival venue is excluded, except NeurIPS 2026 papers through fast-track consideration.\n\nReview Policy: Double-blind review process. Every submission requires a short statement covering potential societal impacts and suggested mitigations.\n\nPresentation: Accepted papers appear as posters, with some selected for oral or spotlight presentations.\n\nCurrent submission requirements are tentative, and additional formats may be added. All deadlines follow Anywhere on Earth (AoE).\n\nImportant Dates:\n- Submission deadline: August 29, 2026, 11:59 PM AoE\n- Author notification: September 29, 2026\n- Workshop event: December 12 or 13, 2026, Atlanta, Georgia, USA" }, { "key": "LP4FM", "title": "NeurIPS 2026 Workshop on Linguistic Principles for Foundation Models", "subtitle": "NeurIPS 2026 Workshop LP4FM", "summary": "A workshop that treats the linguistic and symbolic medium through which foundation models perceive, reason, and act as a first-class design axis, on par with architecture and scale.", "cfp_full": "Linguistic Principles for Foundation Models (LP4FM)\nNeurIPS 2026 | Dec 11 or 12, 2026 | Sydney, Australia\n\nRephrase a problem, and a model's reasoning can collapse. What if language itself is the design axis we've been missing?\n\nWorkshop Overview\nThe scaling paradigm is approaching its limits on reasoning-intensive tasks, and agentic and scientific applications increasingly push models into domains where free-form natural language is demonstrably inadequate. This workshop treats the linguistic and symbolic medium through which foundation models perceive, reason, and act as a first-class design axis, on par with architecture and scale.\n\nCall for Papers\nWe invite high-quality submissions at the intersection of linguistic principles and foundation models. Accepted papers will be presented as posters, and selected submissions will be invited for contributed spotlight talks.\n\nSubmission Guidelines\nSubmissions must be in English and use the NeurIPS 2026 workshop LaTeX template. Papers are submitted as a single PDF:\n- Full Papers: at most 9 pages (main text)\n- Short Papers: at most 4 pages (main text)\n- Demo Track: live demonstrations of systems and tools that put linguistic principles to work, presented alongside the poster sessions (4 pages max in main text)\n- Position Papers: on the role of linguistic and symbolic media in the design, evaluation, and societal impact of foundation models (9 pages max in main text)\nReferences and appendices do not count toward the page limit, but the main text must be self-contained.\nReview: Double-blind. Submissions must be fully anonymized.\nPortal: Managed through OpenReview.\nNon-archival: Authors retain full copyright; extended versions may be submitted elsewhere. Under-review NeurIPS papers are welcome.\n\nImportant Dates\nJul 18, 2026 - Call for Papers opens\nAug 29, 2026 - Submission deadline (Non-archival; under-review NeurIPS papers are welcome and will receive full consideration.)\nSep 1-26, 2026 - Review period\nSep 29, 2026 - Author notification (AoE · NeurIPS mandatory deadline)\nOct 10, 2026 - Camera-ready (AoE)\nDec 11/12, 2026 - Workshop day (NeurIPS 2026 · Sydney, Australia)\n\nTopics of Interest\nTopics include, but are not limited to, the following.\ni. Language Representation Design: Structured, symbolic, or formal representations - including typed slots, intermediate notations, and formal grammars - that shape model schemas and improve reasoning, planning, and tool use.\nii. Compositionality, Semantics, and Pragmatics: How foundation models compose meaning from parts and generalize to novel combinations, and how reference, presupposition, implicature, scope, modality, and discourse structure are handled.\niii. Typology, Multilinguality, Morphology, and Tokenization: Cross-linguistic variation, low-resource languages, language-universal versus language-specific representations, and how tokenization and morphological complexity shape capability, including principled alternatives to current schemes.\niv. Psycholinguistics and Language Acquisition: Foundation models as learners and cognitive models, with comparisons to human acquisition and processing. Bridging developmental psycholinguistics and machine learning.\nv. Probing, Interpretability, and Emergent Linguistic Structure: Mechanistic analysis of how syntactic, semantic, and pragmatic structures are encoded in model internals: how linguistic structure emerges, is represented, and can be recovered from neural computations.\nvi. Reasoning, Agents, and Formal / Scientific Languages: How paraphrases, formal rewrites, and intermediate languages (code, math, logic) reshape reasoning on the same task; structured languages for agent planning and tool use; and domain-specific notations such as Lean, PDDL, and LTL as media of model reasoning.\nvii. Linguistically Grounded Evaluation: Benchmarks beyond surface accuracy, covering compositional generalization, paraphrase robustness, pragmatic inference, and cross-linguistic competence.\nviii. Beyond Text, Self-Designed Languages, and Theory: Extending linguistic principles to vision-language, speech, sign-language, and embodied foundation models; models that design their own symbolic media; and theoretical frameworks linking language structure, schema formation, and model capability.\n\nThe workshop is held in-person at NeurIPS 2026 in Sydney. Remote presentations are supported to accommodate last-minute schedule changes for invited speakers and authors. The workshop proceedings are non-archival. Previously published work is welcome for presentation but is not eligible for workshop awards.", "cfp_status": "published", "topics": [ "Language Representation Design", "Compositionality, Semantics, and Pragmatics", "Typology, Multilinguality, Morphology, and Tokenization", "Psycholinguistics and Language Acquisition", "Probing, Interpretability, and Emergent Linguistic Structure", "Reasoning, Agents, and Formal / Scientific Languages", "Linguistically Grounded Evaluation", "Beyond Text, Self-Designed Languages, and Theory" ], "important_dates": [ { "label": "Call for Papers opens", "date": "2026-07-18" }, { "label": "Submission deadline", "date": "2026-08-29" }, { "label": "Review period", "date": "2026-09-01 to 2026-09-26" }, { "label": "Author notification", "date": "2026-09-29" }, { "label": "Camera-ready", "date": "2026-10-10" }, { "label": "Workshop day", "date": "2026-12-11" } ], "organizers": [ "Zhiqin Yang (Hong Kong University of Science and Technology)", "Yuhan Liu (Xiaomi Inc.)", "Xiuying Chen (Mohamed bin Zayed University of Artificial Intelligence)", "Ekaterina Vylomova (University of Melbourne)", "Bo Han (Hong Kong Baptist University & RIKEN AIP)", "Masashi Sugiyama (RIKEN AIP & University of Tokyo)", "Jingwen Fu (Zhongguancun Academy)", "Yike Guo (The Hong Kong University of Science and Technology, Advisory Panel)" ], "speakers": [ "Roger Levy (Massachusetts Institute of Technology)", "Mirella Lapata (University of Edinburgh)", "Gholamreza Haffari (Monash University & OpenStream AI)", "Jie Fu (IQuest Research)", "Tal Linzen (New York University & Google Research)", "Nouha Dziri (Cohere Labs)", "David Ifeoluwa Adelani (McGill University & Mila)", "Jey Han Lau (University of Melbourne)" ], "host_url": "https://lp4fm.github.io/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/LP4FM", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/LP4FM", "location": "Sydney, Australia", "city": "Sydney", "workshop_date": "2026-07-14", "contact": "yangzqccc@gmail.com", "tracks": [ { "key": "LP4FM", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/LP4FM", "submission_dates_raw": "" } ], "group": "fm", "group_label": "Foundation Models & LLMs", "corpus": "NeurIPS 2026 Workshop on Linguistic Principles for Foundation Models NeurIPS 2026 Workshop LP4FM A workshop that treats the linguistic and symbolic medium through which foundation models perceive, reason, and act as a first-class design axis, on par with architecture and scale. Language Representation Design Compositionality, Semantics, and Pragmatics Typology, Multilinguality, Morphology, and Tokenization Psycholinguistics and Language Acquisition Probing, Interpretability, and Emergent Linguistic Structure Reasoning, Agents, and Formal / Scientific Languages Linguistically Grounded Evaluation Beyond Text, Self-Designed Languages, and Theory Linguistic Principles for Foundation Models (LP4FM)\nNeurIPS 2026 | Dec 11 or 12, 2026 | Sydney, Australia\n\nRephrase a problem, and a model's reasoning can collapse. What if language itself is the design axis we've been missing?\n\nWorkshop Overview\nThe scaling paradigm is approaching its limits on reasoning-intensive tasks, and agentic and scientific applications increasingly push models into domains where free-form natural language is demonstrably inadequate. This workshop treats the linguistic and symbolic medium through which foundation models perceive, reason, and act as a first-class design axis, on par with architecture and scale.\n\nCall for Papers\nWe invite high-quality submissions at the intersection of linguistic principles and foundation models. Accepted papers will be presented as posters, and selected submissions will be invited for contributed spotlight talks.\n\nSubmission Guidelines\nSubmissions must be in English and use the NeurIPS 2026 workshop LaTeX template. Papers are submitted as a single PDF:\n- Full Papers: at most 9 pages (main text)\n- Short Papers: at most 4 pages (main text)\n- Demo Track: live demonstrations of systems and tools that put linguistic principles to work, presented alongside the poster sessions (4 pages max in main text)\n- Position Papers: on the role of linguistic and symbolic media in the design, evaluation, and societal impact of foundation models (9 pages max in main text)\nReferences and appendices do not count toward the page limit, but the main text must be self-contained.\nReview: Double-blind. Submissions must be fully anonymized.\nPortal: Managed through OpenReview.\nNon-archival: Authors retain full copyright; extended versions may be submitted elsewhere. Under-review NeurIPS papers are welcome.\n\nImportant Dates\nJul 18, 2026 - Call for Papers opens\nAug 29, 2026 - Submission deadline (Non-archival; under-review NeurIPS papers are welcome and will receive full consideration.)\nSep 1-26, 2026 - Review period\nSep 29, 2026 - Author notification (AoE · NeurIPS mandatory deadline)\nOct 10, 2026 - Camera-ready (AoE)\nDec 11/12, 2026 - Workshop day (NeurIPS 2026 · Sydney, Australia)\n\nTopics of Interest\nTopics include, but are not limited to, the following.\ni. Language Representation Design: Structured, symbolic, or formal representations - including typed slots, intermediate notations, and formal grammars - that shape model schemas and improve reasoning, planning, and tool use.\nii. Compositionality, Semantics, and Pragmatics: How foundation models compose meaning from parts and generalize to novel combinations, and how reference, presupposition, implicature, scope, modality, and discourse structure are handled.\niii. Typology, Multilinguality, Morphology, and Tokenization: Cross-linguistic variation, low-resource languages, language-universal versus language-specific representations, and how tokenization and morphological complexity shape capability, including principled alternatives to current schemes.\niv. Psycholinguistics and Language Acquisition: Foundation models as learners and cognitive models, with comparisons to human acquisition and processing. Bridging developmental psycholinguistics and machine learning.\nv. Probing, Interpretability, and Emergent Linguistic Structure: Mechanistic analysis of how syntactic, semantic, and pragmatic structures are encoded in model internals: how linguistic structure emerges, is represented, and can be recovered from neural computations.\nvi. Reasoning, Agents, and Formal / Scientific Languages: How paraphrases, formal rewrites, and intermediate languages (code, math, logic) reshape reasoning on the same task; structured languages for agent planning and tool use; and domain-specific notations such as Lean, PDDL, and LTL as media of model reasoning.\nvii. Linguistically Grounded Evaluation: Benchmarks beyond surface accuracy, covering compositional generalization, paraphrase robustness, pragmatic inference, and cross-linguistic competence.\nviii. Beyond Text, Self-Designed Languages, and Theory: Extending linguistic principles to vision-language, speech, sign-language, and embodied foundation models; models that design their own symbolic media; and theoretical frameworks linking language structure, schema formation, and model capability.\n\nThe workshop is held in-person at NeurIPS 2026 in Sydney. Remote presentations are supported to accommodate last-minute schedule changes for invited speakers and authors. The workshop proceedings are non-archival. Previously published work is welcome for presentation but is not eligible for workshop awards." }, { "key": "Med-Reasoner", "title": "NeurIPS 2026 Workshop on Medical Reasoning with Vision Language Foundation Models", "subtitle": "Med-Reasoner 2026 NeurIPS", "summary": "A workshop on reasoning capabilities of vision-language foundation models for medical imaging and clinical decision-making, bringing together computer vision researchers, medical AI experts, imaging scientists, and clinicians to advance interpretable, trustworthy medical AI.", "cfp_full": "Introduction\n\nVision-language foundation models demonstrate strong performance in pattern recognition, but lack robust reasoning capabilities, limiting their application in specialized domains. Medical imaging highlights this critical limitation: diagnosis requires connecting visual findings with clinical knowledge through explicit reasoning processes. While foundation models excel at visual recognition, medical decision-making demands interpretable chain-of-thought diagnosis, multimodal grounding of visual features in clinical knowledge, comprehensive evaluation frameworks for reasoning quality, and probabilistic reasoning for clinical decisions. Given the rapid advancements in foundation models over the past three years, addressing these reasoning gaps has become essential for deploying AI systems in high-stakes healthcare applications where explainability and trustworthiness are paramount.\n\nThe NeurIPS 2026 Workshop on Medical Reasoning with Vision Language Foundation Models (Med-Reasoner) aims to bring together computer vision researchers, medical AI experts, imaging scientists, and practicing clinicians to discuss state-of-the-art advancements, applications, and challenges in reasoning capabilities for medical vision-language models. The workshop will foster discussions that inspire innovation in interpretable medical AI and address real-world deployment challenges including privacy constraints, workflow integration into healthcare systems, and ensuring fairness across patient populations.\n\nScope\n\nThe workshop invites submissions of original research papers exploring the development, application, and evaluation of reasoning capabilities in vision-language foundation models for medical imaging and clinical decision-making. Accepted papers are non-archival and will be presented without appearing in proceedings.\n\nTopics of Interest\n\nReasoning Architectures:\n- Chain-of-Thought Reasoning for Medical Diagnosis\n- Interpretable and Explainable Decision-Making in Medical AI\n- Multimodal Reasoning Across Imaging, Clinical Notes, and Pathology\n- Probabilistic Reasoning Frameworks for Clinical Decisions\n- Grounding Visual Features in Clinical Knowledge\n\nEvaluation and Benchmarking:\n- Evaluation Frameworks for Reasoning Quality Assessment\n- Metrics for Clinical Reasoning Validation\n- Benchmarks for Diagnostic Reasoning in Medical Imaging\n- Human-AI Collaboration and Clinical Validation Studies\n\nFoundation Models:\n- Vision-Language Models for Radiology, Pathology, and Clinical Imaging\n- Multimodal Medical Foundation Models\n- Few-Shot and Zero-Shot Reasoning in Medical Contexts\n- Retrieval-Augmented Medical Reasoning Systems\n\nClinical Deployment and Safety:\n- Bias Detection and Fairness Across Patient Populations\n- Privacy-Preserving Reasoning Systems\n- Workflow Integration in Healthcare Settings\n- Hallucination Detection and Mitigation in Medical Explanations\n\nEmerging Topics:\n- Large Language Models for Clinical Reasoning\n- Agent-Based Medical AI Systems\n- Interpretability and Transparency in Medical Foundation Models\n- Ethical Aspects of Reasoning-Based Medical AI\n\nSubmission Guidelines\n\nFormat: Long papers are limited to eight pages and short papers to four pages, including figures/tables, using the NeurIPS 2026 template. References may exceed page limits.\n\nReview Process: Double-blind peer review conducted through OpenReview with limited visibility to assigned reviewers only.\n\nEthics Requirements: All submissions must disclose dataset licensing and governance protocols while ensuring protected health information is excluded.\n\nImportant Dates\n\nSubmission Deadline: August 22, 2026, 11:59 PM AoE\nAccept / Reject Notification: September 29, 2026\nWorkshop Date: December 12-13, 2026, 8:00 AM - 5:00 PM, Atlanta, Georgia, USA", "cfp_status": "published", "topics": [ "Chain-of-Thought Reasoning for Medical Diagnosis", "Interpretable and Explainable Decision-Making in Medical AI", "Multimodal Reasoning Across Imaging, Clinical Notes, and Pathology", "Probabilistic Reasoning Frameworks for Clinical Decisions", "Grounding Visual Features in Clinical Knowledge", "Evaluation Frameworks for Reasoning Quality Assessment", "Metrics for Clinical Reasoning Validation", "Benchmarks for Diagnostic Reasoning in Medical Imaging", "Human-AI Collaboration and Clinical Validation Studies", "Vision-Language Models for Radiology, Pathology, and Clinical Imaging", "Multimodal Medical Foundation Models", "Few-Shot and Zero-Shot Reasoning in Medical Contexts", "Retrieval-Augmented Medical Reasoning Systems", "Bias Detection and Fairness Across Patient Populations", "Privacy-Preserving Reasoning Systems", "Workflow Integration in Healthcare Settings", "Hallucination Detection and Mitigation in Medical Explanations", "Large Language Models for Clinical Reasoning", "Agent-Based Medical AI Systems", "Interpretability and Transparency in Medical Foundation Models", "Ethical Aspects of Reasoning-Based Medical AI" ], "important_dates": [ { "label": "Submission Deadline", "date": "2026-08-22" }, { "label": "Accept / Reject Notification", "date": "2026-09-29" }, { "label": "Workshop Date", "date": "2026-12-12" } ], "organizers": [ "Anas Zafar (MD Anderson Cancer Center)", "Sana Tonekaboni (Broad Institute of MIT and Harvard)", "Min Woo Sun (Stanford University)", "Julia Vogt (ETH Zurich)", "Jia Wu (MD Anderson Cancer Center)", "Alejandro Lozano (Stanford University)" ], "speakers": [ "Jason Fries (Stanford University)", "Mike Schaekermann (Google Health)", "Shalmali Joshi (Columbia University)", "Michael Moor (ETH Zurich)", "Sam Schmidgall (Johns Hopkins University)", "Muhammad Mamdani (University of Toronto / University Health Network)" ], "host_url": "https://med-reasoner.github.io/neurips2026/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/Med-Reasoner", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/Med-Reasoner", "location": "Atlanta, Georgia, USA", "city": "Atlanta", "workshop_date": "2026-12-13", "contact": "anaszafar98@gmail.com", "tracks": [ { "key": "Med-Reasoner", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/Med-Reasoner", "submission_dates_raw": "" } ], "group": "health", "group_label": "Health & Medicine", "corpus": "NeurIPS 2026 Workshop on Medical Reasoning with Vision Language Foundation Models Med-Reasoner 2026 NeurIPS A workshop on reasoning capabilities of vision-language foundation models for medical imaging and clinical decision-making, bringing together computer vision researchers, medical AI experts, imaging scientists, and clinicians to advance interpretable, trustworthy medical AI. Chain-of-Thought Reasoning for Medical Diagnosis Interpretable and Explainable Decision-Making in Medical AI Multimodal Reasoning Across Imaging, Clinical Notes, and Pathology Probabilistic Reasoning Frameworks for Clinical Decisions Grounding Visual Features in Clinical Knowledge Evaluation Frameworks for Reasoning Quality Assessment Metrics for Clinical Reasoning Validation Benchmarks for Diagnostic Reasoning in Medical Imaging Human-AI Collaboration and Clinical Validation Studies Vision-Language Models for Radiology, Pathology, and Clinical Imaging Multimodal Medical Foundation Models Few-Shot and Zero-Shot Reasoning in Medical Contexts Retrieval-Augmented Medical Reasoning Systems Bias Detection and Fairness Across Patient Populations Privacy-Preserving Reasoning Systems Workflow Integration in Healthcare Settings Hallucination Detection and Mitigation in Medical Explanations Large Language Models for Clinical Reasoning Agent-Based Medical AI Systems Interpretability and Transparency in Medical Foundation Models Ethical Aspects of Reasoning-Based Medical AI Introduction\n\nVision-language foundation models demonstrate strong performance in pattern recognition, but lack robust reasoning capabilities, limiting their application in specialized domains. Medical imaging highlights this critical limitation: diagnosis requires connecting visual findings with clinical knowledge through explicit reasoning processes. While foundation models excel at visual recognition, medical decision-making demands interpretable chain-of-thought diagnosis, multimodal grounding of visual features in clinical knowledge, comprehensive evaluation frameworks for reasoning quality, and probabilistic reasoning for clinical decisions. Given the rapid advancements in foundation models over the past three years, addressing these reasoning gaps has become essential for deploying AI systems in high-stakes healthcare applications where explainability and trustworthiness are paramount.\n\nThe NeurIPS 2026 Workshop on Medical Reasoning with Vision Language Foundation Models (Med-Reasoner) aims to bring together computer vision researchers, medical AI experts, imaging scientists, and practicing clinicians to discuss state-of-the-art advancements, applications, and challenges in reasoning capabilities for medical vision-language models. The workshop will foster discussions that inspire innovation in interpretable medical AI and address real-world deployment challenges including privacy constraints, workflow integration into healthcare systems, and ensuring fairness across patient populations.\n\nScope\n\nThe workshop invites submissions of original research papers exploring the development, application, and evaluation of reasoning capabilities in vision-language foundation models for medical imaging and clinical decision-making. Accepted papers are non-archival and will be presented without appearing in proceedings.\n\nTopics of Interest\n\nReasoning Architectures:\n- Chain-of-Thought Reasoning for Medical Diagnosis\n- Interpretable and Explainable Decision-Making in Medical AI\n- Multimodal Reasoning Across Imaging, Clinical Notes, and Pathology\n- Probabilistic Reasoning Frameworks for Clinical Decisions\n- Grounding Visual Features in Clinical Knowledge\n\nEvaluation and Benchmarking:\n- Evaluation Frameworks for Reasoning Quality Assessment\n- Metrics for Clinical Reasoning Validation\n- Benchmarks for Diagnostic Reasoning in Medical Imaging\n- Human-AI Collaboration and Clinical Validation Studies\n\nFoundation Models:\n- Vision-Language Models for Radiology, Pathology, and Clinical Imaging\n- Multimodal Medical Foundation Models\n- Few-Shot and Zero-Shot Reasoning in Medical Contexts\n- Retrieval-Augmented Medical Reasoning Systems\n\nClinical Deployment and Safety:\n- Bias Detection and Fairness Across Patient Populations\n- Privacy-Preserving Reasoning Systems\n- Workflow Integration in Healthcare Settings\n- Hallucination Detection and Mitigation in Medical Explanations\n\nEmerging Topics:\n- Large Language Models for Clinical Reasoning\n- Agent-Based Medical AI Systems\n- Interpretability and Transparency in Medical Foundation Models\n- Ethical Aspects of Reasoning-Based Medical AI\n\nSubmission Guidelines\n\nFormat: Long papers are limited to eight pages and short papers to four pages, including figures/tables, using the NeurIPS 2026 template. References may exceed page limits.\n\nReview Process: Double-blind peer review conducted through OpenReview with limited visibility to assigned reviewers only.\n\nEthics Requirements: All submissions must disclose dataset licensing and governance protocols while ensuring protected health information is excluded.\n\nImportant Dates\n\nSubmission Deadline: August 22, 2026, 11:59 PM AoE\nAccept / Reject Notification: September 29, 2026\nWorkshop Date: December 12-13, 2026, 8:00 AM - 5:00 PM, Atlanta, Georgia, USA" }, { "key": "PhysUnderstand", "title": "NeurIPS 2026 Workshop on Physical Understanding for Decision-Making: Bridging Foundation Models and Reliable Agents", "subtitle": "NeurIPS 2026 Workshop PhysUnderstand", "summary": "This workshop treats physical understanding as the missing bridge from foundation and world models to reliable decision-making agents, asking how AI systems can understand the physical world well enough to make safe, reliable decisions within it.", "cfp_full": "Physical Understanding for Decision-Making: Bridging Foundation Models and Reliable Agents\n\nPhysical understanding is the missing bridge from foundation and world models to reliable decision-making agents.\n\nPaper Submission Deadline: August 26th, 2026, 08:00 UTC\n\nWorkshop Scope\n\nAI systems are increasingly deployed in physical settings such as robotics, autonomous driving, and laboratory automation. In these settings, actions must respect physical laws. Objects have mass, contacts transmit forces, constraints limit feasibility, and interventions produce irreversible consequences. An agent that lacks physical understanding (the ability to reason about objects, forces, affordances, causal mechanisms, and long-horizon interaction) cannot act safely or reliably, no matter how accurately it predicts the next video frame.\n\nFoundation models for decision-making, including vision-language-action models, robot foundation models, agentic systems, and interactive world models, have sharply improved long-horizon video generation, embodied policy learning, and multimodal action grounding. Yet most of these models are still trained and evaluated primarily for visual realism or short-term prediction accuracy. The field now stands at an inflection point. Foundation and world models are powerful enough to serve as the backbone of physical agents, yet the community lacks shared definitions, evaluation protocols, and benchmarks for physical understanding as a distinct and measurable capability.\n\nThis workshop brings together researchers from machine learning, reinforcement learning, robotics, computer vision, simulation, causality, embodied AI, autonomous driving, and safety around one central question: how can AI systems understand the physical world well enough to make reliable decisions within it?\n\nWe welcome theoretical, algorithmic, empirical, benchmark, systems, and position contributions across four themes.\n\n- Modeling Physical Systems. Object- and scene-centric dynamics, contact, friction, fluids, deformable objects, material properties, 3D and 4D world models, neural simulation, and video-to-physics. Models that capture actionable physical structure rather than appearance alone.\n- Physical Perception and Active Interaction. Acting to reduce uncertainty, learning affordances, and discovering physical constraints under partial observability. How agents can use interaction to build richer physical representations.\n- Causal and Counterfactual Physical Reasoning. Interventions, mechanisms, counterfactual outcomes, stability analysis, and distribution shift. Can models reason about what would happen under actions they have not yet taken?\n- Decision-Making, Evaluation, and Deployment. Planning, model-based RL, control, manipulation, navigation, and long-horizon execution. Evaluation methodology covering controllability, calibration, causal validity, sim-to-real transfer, and robustness, plus deployment concerns such as hallucinated actions, constraint satisfaction, risk estimation, and failure recovery.", "cfp_status": "published", "topics": [ "Modeling Physical Systems (object- and scene-centric dynamics, contact, friction, fluids, deformable objects, material properties, 3D/4D world models, neural simulation, video-to-physics)", "Physical Perception and Active Interaction (acting to reduce uncertainty, learning affordances, discovering physical constraints under partial observability)", "Causal and Counterfactual Physical Reasoning (interventions, mechanisms, counterfactual outcomes, stability analysis, distribution shift)", "Decision-Making, Evaluation, and Deployment (planning, model-based RL, control, manipulation, navigation, long-horizon execution; evaluation of controllability, calibration, causal validity, sim-to-real transfer, robustness; hallucinated actions, constraint satisfaction, risk estimation, failure recovery)" ], "important_dates": [ { "label": "Paper Submission Deadline", "date": "2026-08-26" }, { "label": "Workshop date", "date": "2026-12-11" } ], "organizers": [ "Felix Juefei-Xu (Google DeepMind)", "Tianyu Shi (McGill University)", "Shirley Rong Zou (Apple)", "Mengyue Yang (University of Bristol)", "Aaron Xuxiang Tian (Toyota, Publicity Chair)", "Zhenyu Zhang (OpenAI, Communication Chair)" ], "speakers": [ "Saining Xie (New York University & AMI Labs)", "Lu Lu (Yale University)", "Sergey Levine (UC Berkeley)", "Danijar Hafner (Google DeepMind)", "Shanghang Zhang (Peking University)", "Jiajun Wu (Stanford University)", "Xiaolong Wang (UC San Diego)", "Shuang Li (Google DeepMind)", "Georgia Chalvatzaki (TU Darmstadt)" ], "host_url": "https://sites.google.com/view/neurips-2026-workshop-pudm", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/PhysUnderstand", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/PhysUnderstand", "location": "Sydney", "city": "Sydney", "workshop_date": "2026-12-10", "contact": "physicalunderstanding.workshop@gmail.com", "tracks": [ { "key": "PhysUnderstand", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/PhysUnderstand", "submission_dates_raw": "" } ], "group": "science", "group_label": "AI for Science", "corpus": "NeurIPS 2026 Workshop on Physical Understanding for Decision-Making: Bridging Foundation Models and Reliable Agents NeurIPS 2026 Workshop PhysUnderstand This workshop treats physical understanding as the missing bridge from foundation and world models to reliable decision-making agents, asking how AI systems can understand the physical world well enough to make safe, reliable decisions within it. Modeling Physical Systems (object- and scene-centric dynamics, contact, friction, fluids, deformable objects, material properties, 3D/4D world models, neural simulation, video-to-physics) Physical Perception and Active Interaction (acting to reduce uncertainty, learning affordances, discovering physical constraints under partial observability) Causal and Counterfactual Physical Reasoning (interventions, mechanisms, counterfactual outcomes, stability analysis, distribution shift) Decision-Making, Evaluation, and Deployment (planning, model-based RL, control, manipulation, navigation, long-horizon execution; evaluation of controllability, calibration, causal validity, sim-to-real transfer, robustness; hallucinated actions, constraint satisfaction, risk estimation, failure recovery) Physical Understanding for Decision-Making: Bridging Foundation Models and Reliable Agents\n\nPhysical understanding is the missing bridge from foundation and world models to reliable decision-making agents.\n\nPaper Submission Deadline: August 26th, 2026, 08:00 UTC\n\nWorkshop Scope\n\nAI systems are increasingly deployed in physical settings such as robotics, autonomous driving, and laboratory automation. In these settings, actions must respect physical laws. Objects have mass, contacts transmit forces, constraints limit feasibility, and interventions produce irreversible consequences. An agent that lacks physical understanding (the ability to reason about objects, forces, affordances, causal mechanisms, and long-horizon interaction) cannot act safely or reliably, no matter how accurately it predicts the next video frame.\n\nFoundation models for decision-making, including vision-language-action models, robot foundation models, agentic systems, and interactive world models, have sharply improved long-horizon video generation, embodied policy learning, and multimodal action grounding. Yet most of these models are still trained and evaluated primarily for visual realism or short-term prediction accuracy. The field now stands at an inflection point. Foundation and world models are powerful enough to serve as the backbone of physical agents, yet the community lacks shared definitions, evaluation protocols, and benchmarks for physical understanding as a distinct and measurable capability.\n\nThis workshop brings together researchers from machine learning, reinforcement learning, robotics, computer vision, simulation, causality, embodied AI, autonomous driving, and safety around one central question: how can AI systems understand the physical world well enough to make reliable decisions within it?\n\nWe welcome theoretical, algorithmic, empirical, benchmark, systems, and position contributions across four themes.\n\n- Modeling Physical Systems. Object- and scene-centric dynamics, contact, friction, fluids, deformable objects, material properties, 3D and 4D world models, neural simulation, and video-to-physics. Models that capture actionable physical structure rather than appearance alone.\n- Physical Perception and Active Interaction. Acting to reduce uncertainty, learning affordances, and discovering physical constraints under partial observability. How agents can use interaction to build richer physical representations.\n- Causal and Counterfactual Physical Reasoning. Interventions, mechanisms, counterfactual outcomes, stability analysis, and distribution shift. Can models reason about what would happen under actions they have not yet taken?\n- Decision-Making, Evaluation, and Deployment. Planning, model-based RL, control, manipulation, navigation, and long-horizon execution. Evaluation methodology covering controllability, calibration, causal validity, sim-to-real transfer, and robustness, plus deployment concerns such as hallucinated actions, constraint satisfaction, risk estimation, and failure recovery." }, { "key": "PriGM", "title": "NeurIPS 2026 Workshop on Principles of Generative Modeling (PriGM)", "subtitle": "PriGM NeurIPS2026", "summary": "The 2nd Workshop on Principles of Generative Modeling brings together theory-oriented communities to articulate and synthesize the core principles underlying modern generative AI and to outline central challenges in advancing scientific understanding of generative models.", "cfp_full": "2nd Workshop on Principles of Generative Modeling (PriGM) @ NeurIPS2026 Paris\n\nWorkshop summary\nMachine learning theory has long focused on classical supervised learning settings, where a model is trained on input–label pairs drawn from a well-defined data distribution, with the aim of achieving low test error on such distribution. Despite remarkable advances in such settings, recent breakthroughs in generative AI have transformed our understanding of generalization, revealing phenomena such as emergent capabilities and in-context learning, which lie beyond the scope of existing theoretical frameworks. These empirical developments call for new theoretical paradigms, fostering closer interactions between theoreticians and practitioners to address the distinctive challenges posed by generative models.\n\nThe purpose of this workshop is to bring together diverse theory-oriented communities to articulate and synthesize core principles underlying modern generative AI, and to outline the central challenges in advancing our scientific understanding.\n\nCall for Papers\nWe invite submissions that take principled approaches to advancing the understanding of generative modeling. This understanding may draw from mathematical theory, physical modeling, and rigorous empirical analysis.\n\nTopics of Interest\nThe workshop encourages contributions addressing four foundational areas:\n- Model classes and expressivity: Examining what functions, distributions, or algorithms modern generative models can represent and how architectural choices influence expressive power. This includes investigation of transformers, diffusion models, discrete diffusions, and state-space models' effectiveness with linguistic or visual structures, plus how repeated token-level computations expand reasoning capabilities.\n- Learning, generalization and inductive bias: Exploring how generative models develop high-level structure and capabilities like in-context learning, compositional generalization, and reasoning. The area investigates unifying principles across architectures and modalities versus architecture-specific inductive biases, plus fundamental limits of self-supervised learning paradigms such as next-token prediction.\n- Inference-time computation and adaptation: Analyzing when and why additional inference-time computation enhances generation quality and reasoning, including sampling and test-time adaptation. This covers statistical and computational limits for adapting models to distribution shifts.\n- Post-training: Examining which pretrained model properties enable successful post-training, whether post-training creates, reveals, or reweights capabilities, and distinctions between supervised fine-tuning, reinforcement learning, and distillation.\n\nSubmission Guidelines\nSubmissions are limited to four single-column pages, plus unlimited pages for references and appendices. The review process employs double-blind evaluation; submissions must be anonymized without author names, affiliations, acknowledgements, or identifying information. All submissions occur via OpenReview using standard NeurIPS LaTeX style files. The NeurIPS checklist is optional, and appendices may be included in the main PDF.\n\nThe workshop is non-archival without official proceedings. Dual submissions to other venues are permitted, and unpublished or under-review work is welcomed. However, papers already accepted elsewhere with archival proceedings cannot be submitted. This is an in-person event requiring poster presentations from authors; a subset will be selected for 15-minute contributed talks.\n\nImportant Dates\n- Paper submission deadline: September 5th, 2026, AoE\n- Review period: September 6th - 22nd\n- Notification date: September 29th, 2026, AoE\n\nSchedule & practical information\n- Date: December 12 or 13 (TBC), 2026\n- Location: Palais des Congrès, Paris, France\n- Poster size: A0 portrait or A1 landscape\n- Contact: prigm-neurips-2026@googlegroups.com", "cfp_status": "published", "topics": [ "Model classes and expressivity", "Learning, generalization and inductive bias", "Inference-time computation and adaptation", "Post-training" ], "important_dates": [ { "label": "Paper submission deadline", "date": "2026-09-05" }, { "label": "Review period", "date": "September 6th - 22nd, 2026" }, { "label": "Notification date", "date": "2026-09-29" }, { "label": "Workshop date", "date": "December 12 or 13, 2026 (TBC)" } ], "organizers": [ "Francesco Cagnetta (SISSA)", "Elisabetta Cornacchia (Bocconi University)", "Soon Hoe Lim (KTH & Nordita)", "Bingbin Liu (Kempner Institute, Harvard)", "Bruno Loureiro (CNRS & École Normale Supérieure)", "Valentin De Bortoli (Google DeepMind)", "Gabriel Peyré (CNRS & École Normale Supérieure)" ], "speakers": [ "Emanuel Abbe (Apple & EPFL)", "Francis Bach (INRIA & École Normale Supérieure)", "Julia Kempe (META & NYU)", "Noam Levi (META)", "Gabriele Steidl (TU Berlin)" ], "host_url": "https://sites.google.com/view/prigmneurips2026/home", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/PriGM", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/PriGM", "location": "Paris, France", "city": "Paris", "workshop_date": "2026-12-11", "contact": "prigm-neurips-2026@googlegroups.com", "tracks": [ { "key": "PriGM", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/PriGM", "submission_dates_raw": "" } ], "group": "theory", "group_label": "Theory & Methods", "corpus": "NeurIPS 2026 Workshop on Principles of Generative Modeling (PriGM) PriGM NeurIPS2026 The 2nd Workshop on Principles of Generative Modeling brings together theory-oriented communities to articulate and synthesize the core principles underlying modern generative AI and to outline central challenges in advancing scientific understanding of generative models. Model classes and expressivity Learning, generalization and inductive bias Inference-time computation and adaptation Post-training 2nd Workshop on Principles of Generative Modeling (PriGM) @ NeurIPS2026 Paris\n\nWorkshop summary\nMachine learning theory has long focused on classical supervised learning settings, where a model is trained on input–label pairs drawn from a well-defined data distribution, with the aim of achieving low test error on such distribution. Despite remarkable advances in such settings, recent breakthroughs in generative AI have transformed our understanding of generalization, revealing phenomena such as emergent capabilities and in-context learning, which lie beyond the scope of existing theoretical frameworks. These empirical developments call for new theoretical paradigms, fostering closer interactions between theoreticians and practitioners to address the distinctive challenges posed by generative models.\n\nThe purpose of this workshop is to bring together diverse theory-oriented communities to articulate and synthesize core principles underlying modern generative AI, and to outline the central challenges in advancing our scientific understanding.\n\nCall for Papers\nWe invite submissions that take principled approaches to advancing the understanding of generative modeling. This understanding may draw from mathematical theory, physical modeling, and rigorous empirical analysis.\n\nTopics of Interest\nThe workshop encourages contributions addressing four foundational areas:\n- Model classes and expressivity: Examining what functions, distributions, or algorithms modern generative models can represent and how architectural choices influence expressive power. This includes investigation of transformers, diffusion models, discrete diffusions, and state-space models' effectiveness with linguistic or visual structures, plus how repeated token-level computations expand reasoning capabilities.\n- Learning, generalization and inductive bias: Exploring how generative models develop high-level structure and capabilities like in-context learning, compositional generalization, and reasoning. The area investigates unifying principles across architectures and modalities versus architecture-specific inductive biases, plus fundamental limits of self-supervised learning paradigms such as next-token prediction.\n- Inference-time computation and adaptation: Analyzing when and why additional inference-time computation enhances generation quality and reasoning, including sampling and test-time adaptation. This covers statistical and computational limits for adapting models to distribution shifts.\n- Post-training: Examining which pretrained model properties enable successful post-training, whether post-training creates, reveals, or reweights capabilities, and distinctions between supervised fine-tuning, reinforcement learning, and distillation.\n\nSubmission Guidelines\nSubmissions are limited to four single-column pages, plus unlimited pages for references and appendices. The review process employs double-blind evaluation; submissions must be anonymized without author names, affiliations, acknowledgements, or identifying information. All submissions occur via OpenReview using standard NeurIPS LaTeX style files. The NeurIPS checklist is optional, and appendices may be included in the main PDF.\n\nThe workshop is non-archival without official proceedings. Dual submissions to other venues are permitted, and unpublished or under-review work is welcomed. However, papers already accepted elsewhere with archival proceedings cannot be submitted. This is an in-person event requiring poster presentations from authors; a subset will be selected for 15-minute contributed talks.\n\nImportant Dates\n- Paper submission deadline: September 5th, 2026, AoE\n- Review period: September 6th - 22nd\n- Notification date: September 29th, 2026, AoE\n\nSchedule & practical information\n- Date: December 12 or 13 (TBC), 2026\n- Location: Palais des Congrès, Paris, France\n- Poster size: A0 portrait or A1 landscape\n- Contact: prigm-neurips-2026@googlegroups.com" }, { "key": "ReMuCAI", "title": "NeurIPS 2026 Workshop on Real-Time Multimodal Conversational AI", "subtitle": "ReMuCAI 2026", "summary": "The first NeurIPS Real-Time Multimodal Conversational AI workshop advances research on agents that perceive scenes and humans through multimodal signals and generate speech and non-verbal communication in real-time, integrating egocentric, exocentric, and dyadic perspectives for seamless human-machine collaboration.", "cfp_full": "Workshop Description\nFor a long time, conversational AI has been confined to unimodal text or speech exchanges. As human-machine dialogue increasingly extends into the physical world through embodied agents and virtual assistants, agents must perceive scenes and humans through multimodal signals (e.g., speech, video, sensor data). To enable seamless human-machine collaboration, they must also generate speech and non-verbal communication, maintain a coherent conversation, and, when necessary, interact with the physical world in real-time, with low latency, and in a context-aware manner.\n\nHistorically, research on these challenges has been fragmented. For example, egocentric conversational AI focuses on interactions from the user's perspective, such as AI assistants or augmented-reality glasses. Conversely, exocentric and dyadic conversational AI centers on third-person perspectives and face-to-face communication, typical of traditional robotics. However, real-time multimodal conversations naturally demand an integration of multiple perspectives to perform cross-view reasoning. For instance, a seamless dialogue about assembling furniture requires an understanding of the user's view (egocentric), the broader context of the room and objects (exocentric), and human emotions, gestures, and motion (dyadic).\n\nThese challenges outline four core research areas that require interdisciplinary collaboration across computer vision, robotics, machine learning, speech and audio processing, and dialogue communities:\n- Multimodal representation and perception\n- Real-time interaction dynamics and memory\n- Embodied interaction\n- Benchmarks and datasets\n\nThe first NeurIPS Real-Time Multimodal Conversational AI workshop aims to advance progress across these critical research areas.\n\nCall for Papers\nThe workshop aims to convene diverse communities discussing real-time multimodal conversational AI, welcoming negative results if relevant alongside positive findings. The organizers discourage submissions on black box systems or minor hyperparameter tuning, but permit papers with incremental technical advances, particularly those emphasizing multimodal dimensions.\n\nResearch Topics Welcome:\n- Multimodal representation learning for conversational AI\n- Full-duplex multimodal conversational models\n- Social intelligence and communication for digital/embodied agents\n- Egocentric multimodal perception\n- Multi-speaker conversation\n- Real-time multimodal processing for interactive systems\n- Human-machine interaction in physical environments\n- Predictive turn taking\n- Speech/text controlled expressive gesture and facial expression synthesis\n- Datasets and benchmarks relevant to multimodal conversational AI\n\nSubmission Requirements\nPapers must follow NeurIPS 2026 paper guidelines. Two submission tracks exist: short papers (4 pages maximum excluding references) presenting work-in-progress, and long papers (9 pages maximum excluding references). This non-archival venue will not publish accepted papers in formal proceedings. Submission portal: OpenReview.\n\nKey Dates\n- Submission deadline: August 29, 2026 (AoE)\n- Acceptance notifications: September 29, 2026 (AoE)\n- Workshop dates: December 12–13, 2026, Paris, France", "cfp_status": "published", "topics": [ "Multimodal representation learning for conversational AI", "Full-duplex multimodal conversational models", "Social intelligence and communication for digital/embodied agents", "Egocentric multimodal perception", "Multi-speaker conversation", "Real-time multimodal processing for interactive systems", "Human-machine interaction in physical environments", "Predictive turn taking", "Speech/text controlled expressive gesture and facial expression synthesis", "Datasets and benchmarks relevant to multimodal conversational AI" ], "important_dates": [ { "label": "Submission Deadline", "date": "2026-08-29" }, { "label": "Acceptance Notification", "date": "2026-09-29" }, { "label": "Workshop", "date": "2026-12-12" } ], "organizers": [], "speakers": [], "host_url": "https://sites.google.com/view/remucai", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/ReMuCAI", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/ReMuCAI", "location": "Paris, France", "city": "Paris", "workshop_date": "2026-12-12", "contact": "remucai-neurips26@googlegroups.com", "tracks": [ { "key": "ReMuCAI", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/ReMuCAI", "submission_dates_raw": "" } ], "group": "agents", "group_label": "Agents & Agentic Systems", "corpus": "NeurIPS 2026 Workshop on Real-Time Multimodal Conversational AI ReMuCAI 2026 The first NeurIPS Real-Time Multimodal Conversational AI workshop advances research on agents that perceive scenes and humans through multimodal signals and generate speech and non-verbal communication in real-time, integrating egocentric, exocentric, and dyadic perspectives for seamless human-machine collaboration. Multimodal representation learning for conversational AI Full-duplex multimodal conversational models Social intelligence and communication for digital/embodied agents Egocentric multimodal perception Multi-speaker conversation Real-time multimodal processing for interactive systems Human-machine interaction in physical environments Predictive turn taking Speech/text controlled expressive gesture and facial expression synthesis Datasets and benchmarks relevant to multimodal conversational AI Workshop Description\nFor a long time, conversational AI has been confined to unimodal text or speech exchanges. As human-machine dialogue increasingly extends into the physical world through embodied agents and virtual assistants, agents must perceive scenes and humans through multimodal signals (e.g., speech, video, sensor data). To enable seamless human-machine collaboration, they must also generate speech and non-verbal communication, maintain a coherent conversation, and, when necessary, interact with the physical world in real-time, with low latency, and in a context-aware manner.\n\nHistorically, research on these challenges has been fragmented. For example, egocentric conversational AI focuses on interactions from the user's perspective, such as AI assistants or augmented-reality glasses. Conversely, exocentric and dyadic conversational AI centers on third-person perspectives and face-to-face communication, typical of traditional robotics. However, real-time multimodal conversations naturally demand an integration of multiple perspectives to perform cross-view reasoning. For instance, a seamless dialogue about assembling furniture requires an understanding of the user's view (egocentric), the broader context of the room and objects (exocentric), and human emotions, gestures, and motion (dyadic).\n\nThese challenges outline four core research areas that require interdisciplinary collaboration across computer vision, robotics, machine learning, speech and audio processing, and dialogue communities:\n- Multimodal representation and perception\n- Real-time interaction dynamics and memory\n- Embodied interaction\n- Benchmarks and datasets\n\nThe first NeurIPS Real-Time Multimodal Conversational AI workshop aims to advance progress across these critical research areas.\n\nCall for Papers\nThe workshop aims to convene diverse communities discussing real-time multimodal conversational AI, welcoming negative results if relevant alongside positive findings. The organizers discourage submissions on black box systems or minor hyperparameter tuning, but permit papers with incremental technical advances, particularly those emphasizing multimodal dimensions.\n\nResearch Topics Welcome:\n- Multimodal representation learning for conversational AI\n- Full-duplex multimodal conversational models\n- Social intelligence and communication for digital/embodied agents\n- Egocentric multimodal perception\n- Multi-speaker conversation\n- Real-time multimodal processing for interactive systems\n- Human-machine interaction in physical environments\n- Predictive turn taking\n- Speech/text controlled expressive gesture and facial expression synthesis\n- Datasets and benchmarks relevant to multimodal conversational AI\n\nSubmission Requirements\nPapers must follow NeurIPS 2026 paper guidelines. Two submission tracks exist: short papers (4 pages maximum excluding references) presenting work-in-progress, and long papers (9 pages maximum excluding references). This non-archival venue will not publish accepted papers in formal proceedings. Submission portal: OpenReview.\n\nKey Dates\n- Submission deadline: August 29, 2026 (AoE)\n- Acceptance notifications: September 29, 2026 (AoE)\n- Workshop dates: December 12–13, 2026, Paris, France" }, { "key": "SaTQuML", "title": "NeurIPS 2026 Workshop on SaTQuML: Secure and Trustworthy Quantum Machine Learning", "subtitle": "NeurIPS 2026 SaTQuML Workshop", "summary": "SaTQuML brings together researchers working at the intersection of quantum machine learning, cybersecurity, trustworthy AI, quantum security, and cyberdefense, focusing on the security and trustworthiness of QML systems and the use of QML and hybrid quantum-classical methods for cyberdefense.", "cfp_full": "SaTQuML: Secure and Trustworthy Quantum Machine Learning\nNeurIPS 2026 Workshop · Atlanta, Georgia, United States · December 12-13, 2026\n\nAbout\nSaTQuML: Secure and Trustworthy Quantum Machine Learning brings together researchers working at the intersection of quantum machine learning (QML), cybersecurity, trustworthy AI, quantum security, and cyberdefense. As QML moves toward practical testing on hybrid quantum-classical platforms, its use in security-sensitive settings must be guided by clear benchmarks, realistic threat models, strong baselines, and careful evaluation.\n\nThe workshop focuses on two connected themes. The first is the security and trustworthiness of QML systems themselves, including robustness, privacy, reliability, interpretability, deployment risks, model leakage, and behavior under noisy hardware conditions. The second is the use of QML and hybrid quantum-classical methods for cyberdefense and related security applications, including anomaly detection, malware analysis, intrusion detection, vulnerability prioritization, cyber-physical security, and threat intelligence.\n\nQML is entering an important transition period. Academic research groups and industry platforms are making hybrid quantum-classical experimentation increasingly accessible through open-source software, simulators, cloud-based access to quantum computers, and early application-driven demonstrations. This creates a need to evaluate QML beyond expressivity, trainability, and quantum-advantage claims, with attention to realistic data, reproducibility, robustness, privacy, security, and deployment constraints.\n\nSaTQuML emphasizes rigorous evaluation over speculative claims. Its central goal is to help define meaningful research problems, strong classical and quantum baselines, realistic threat models, reproducible benchmarks, and shared best practices for secure and trustworthy QML systems in security-critical settings.\n\nKey Problems We Aim to Address\n- Trustworthy QML Systems: Define how quantum machine learning systems should be evaluated for security, robustness, privacy, reliability, interpretability, and model leakage. Contributions may study how QML models behave under noisy hardware, limited data, adversarial manipulation, and distribution shift.\n- Benchmarks and Evaluation: Develop realistic benchmarks, datasets, threat models, and evaluation protocols for secure and trustworthy QML. Contributions may compare QML methods against strong classical, quantum, and quantum-inspired baselines, including negative results and studies showing when QML is useful, limited, or unlikely to help.\n- Cyberdefense Applications: Explore QML and hybrid quantum-classical models for cyberdefense tasks such as intrusion detection, anomaly detection, malware analysis, phishing and fraud detection, vulnerability prioritization, cyber-physical security, and threat intelligence.\n- Secure Deployment: Study how QML pipelines can be securely trained, deployed, and accessed in cloud, edge, quantum-cloud, and hybrid computing environments. Contributions may address deployment risks, quantum-cloud access, secure quantum computing, quantum-network security, and post-quantum cyberdefense.\n\nCall for Papers\nNeurIPS 2026 Workshop on SaTQuML: Secure and Trustworthy Quantum Machine Learning invites submissions from researchers working on secure, reliable, and realistic quantum machine learning. We welcome work on trustworthy QML systems, rigorous benchmarking, cyberdefense applications, and secure deployment.\n\nKey Dates\n- Submission Start: August 15, 2026 (AoE)\n- Submission Deadline: August 29, 2026 (AoE)\n- Acceptance Notification: September 29, 2026 (AoE)\n- Camera-ready Deadline: October 15, 2026 (AoE)\n\nSubmission Site\nSubmit papers through the SaTQuML submission portal on OpenReview.\n\nSubmission Tracks\n- Long Paper: 9 pages.\n- Short Paper: 4 pages.\n- Tiny Paper: 2 pages.\n\nScope\nPrimary themes for the workshop include:\n- Trustworthy QML Systems: Adversarial robustness, privacy, reliability, model leakage, hardware noise, and distribution shift.\n- Rigorous Benchmarks and Evaluation: Realistic datasets and threat models; strong baselines; reproducibility, ablations, and negative results.\n- QML for Cyberdefense: Intrusion, anomaly, malware, fraud, and cyber-physical defense; vulnerability prioritization and threat intelligence.\n- Secure QML Deployment: Secure training and inference across cloud, edge, quantum-cloud, and hybrid systems; remote-access and pipeline risks.\n\nExample subtopics and concrete directions include:\n- Security and trustworthiness of QML systems, including robustness, privacy, fairness, reliability, interpretability, and model leakage.\n- Adversarial evaluation of QML, including perturbation, poisoning, evasion, extraction, quantum noise, and distribution-shift settings.\n- Security-sensitive learning methods, including quantum kernels, variational quantum circuits, quantum neural networks, quantum-inspired learning, and hybrid quantum-classical models.\n- Benchmarking and reproducibility, including realistic datasets, strong classical, quantum, and quantum-inspired baselines, ablations, and hardware-aware evaluation.\n- Cyberdefense applications such as intrusion detection, anomaly detection, malware analysis, phishing and fraud detection, cyber-physical security, and threat intelligence.\n- Practical utility and limitations of QML, including studies identifying when QML is useful, limited, or unlikely to improve over classical methods.\n- Secure deployment of QML pipelines in cloud, edge, quantum-cloud, and hybrid computing environments.\n- Quantum-era security, including post-quantum cyberdefense, quantum-network security, secure quantum computing, and trustworthy AI methods for quantum and hybrid models.\n\nSubmission Guidelines\nFormat: All submissions must be a single PDF file. We accept long papers up to 9 pages, short papers up to 4 pages, and tiny papers up to 2 pages. References and appendices are not included in the page limit, but the main text must be self-contained. Reviewers are not required to read beyond the main text.\nStyle file: You must format your submission using the NeurIPS 2026 LaTeX style file. Please include the references and supplementary materials in the same PDF. The maximum file size for submissions is 50MB. Submissions that violate the NeurIPS style (e.g., by decreasing margins or font sizes) or page limits may be rejected without further review.\nDual-submission and non-archival policy: We welcome ongoing and unpublished work. We will also accept papers that are under review at the time of submission, or that have been recently accepted, provided they do not breach any dual-submission or anonymity policies of those venues. The workshop is a non-archival venue and will not have official proceedings. Workshop submissions can be subsequently or concurrently submitted to other venues.\nVisibility: Submissions and reviews will not be public. Only accepted papers will be made public.\nDouble-blind reviewing: All submissions must be anonymized and may not contain any identifying information that may violate the double-blind reviewing policy. This policy applies to any supplementary or linked material as well, including code. If you are including links to any external material, it is your responsibility to guarantee anonymous browsing. Please do not include acknowledgements at submission time. If you need to cite one of your own papers, you should do so with adequate anonymization to preserve double-blind reviewing. Any papers found to be violating this policy will be rejected.\nPlease be AWARE: OpenReview's moderation policy for newly created profiles in the Call for Papers: New profiles created without an institutional email will go through a moderation process that can take up to two weeks. New profiles created with an institutional email will be activated automatically.\nContact: For any questions, please contact us at SaTQuML@gmail.com", "cfp_status": "published", "topics": [ "Trustworthy QML Systems (adversarial robustness, privacy, reliability, model leakage, hardware noise, distribution shift)", "Rigorous Benchmarks and Evaluation (realistic datasets, threat models, strong baselines, reproducibility, negative results)", "QML for Cyberdefense (intrusion, anomaly, malware, fraud, cyber-physical defense, vulnerability prioritization, threat intelligence)", "Secure QML Deployment (cloud, edge, quantum-cloud, hybrid systems; remote-access and pipeline risks)", "Adversarial evaluation of QML (perturbation, poisoning, evasion, extraction, quantum noise)", "Security-sensitive learning methods (quantum kernels, VQCs, quantum neural networks, hybrid models)", "Practical utility and limitations of QML", "Quantum-era security (post-quantum cyberdefense, quantum-network security, secure quantum computing)" ], "important_dates": [ { "label": "Submission Start", "date": "2026-08-15" }, { "label": "Submission Deadline", "date": "2026-08-29" }, { "label": "Acceptance Notification", "date": "2026-09-29" }, { "label": "Camera-ready Deadline", "date": "2026-10-15" } ], "organizers": [ "Saidur Rahman (University of Texas at El Paso)", "Jakub Szefer (Northwestern University)", "Taqi Raza (University of Massachusetts Amherst)", "Khoa Luu (University of Arkansas)", "Shahrooz Pouryousef (Chalmers University of Technology)" ], "speakers": [ "Swaroop Ghosh (Pennsylvania State University)", "Muhammad Usman (Monash University, Australia)", "Samuel Yen-Chi Chen (Wells Fargo)", "Juan Cruz-Benito (IBM Quantum and IBM Research)", "Soohyun Park (Sookmyung Women's University)" ], "host_url": "https://satquml.github.io/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/SaTQuML", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/SaTQuML", "location": "Atlanta, Georgia, USA.", "city": "Atlanta", "workshop_date": "2026-12-12", "contact": "satquml@gmail.com", "tracks": [ { "key": "SaTQuML", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/SaTQuML", "submission_dates_raw": "" } ], "group": "trust", "group_label": "Trustworthy & Safe AI", "corpus": "NeurIPS 2026 Workshop on SaTQuML: Secure and Trustworthy Quantum Machine Learning NeurIPS 2026 SaTQuML Workshop SaTQuML brings together researchers working at the intersection of quantum machine learning, cybersecurity, trustworthy AI, quantum security, and cyberdefense, focusing on the security and trustworthiness of QML systems and the use of QML and hybrid quantum-classical methods for cyberdefense. Trustworthy QML Systems (adversarial robustness, privacy, reliability, model leakage, hardware noise, distribution shift) Rigorous Benchmarks and Evaluation (realistic datasets, threat models, strong baselines, reproducibility, negative results) QML for Cyberdefense (intrusion, anomaly, malware, fraud, cyber-physical defense, vulnerability prioritization, threat intelligence) Secure QML Deployment (cloud, edge, quantum-cloud, hybrid systems; remote-access and pipeline risks) Adversarial evaluation of QML (perturbation, poisoning, evasion, extraction, quantum noise) Security-sensitive learning methods (quantum kernels, VQCs, quantum neural networks, hybrid models) Practical utility and limitations of QML Quantum-era security (post-quantum cyberdefense, quantum-network security, secure quantum computing) SaTQuML: Secure and Trustworthy Quantum Machine Learning\nNeurIPS 2026 Workshop · Atlanta, Georgia, United States · December 12-13, 2026\n\nAbout\nSaTQuML: Secure and Trustworthy Quantum Machine Learning brings together researchers working at the intersection of quantum machine learning (QML), cybersecurity, trustworthy AI, quantum security, and cyberdefense. As QML moves toward practical testing on hybrid quantum-classical platforms, its use in security-sensitive settings must be guided by clear benchmarks, realistic threat models, strong baselines, and careful evaluation.\n\nThe workshop focuses on two connected themes. The first is the security and trustworthiness of QML systems themselves, including robustness, privacy, reliability, interpretability, deployment risks, model leakage, and behavior under noisy hardware conditions. The second is the use of QML and hybrid quantum-classical methods for cyberdefense and related security applications, including anomaly detection, malware analysis, intrusion detection, vulnerability prioritization, cyber-physical security, and threat intelligence.\n\nQML is entering an important transition period. Academic research groups and industry platforms are making hybrid quantum-classical experimentation increasingly accessible through open-source software, simulators, cloud-based access to quantum computers, and early application-driven demonstrations. This creates a need to evaluate QML beyond expressivity, trainability, and quantum-advantage claims, with attention to realistic data, reproducibility, robustness, privacy, security, and deployment constraints.\n\nSaTQuML emphasizes rigorous evaluation over speculative claims. Its central goal is to help define meaningful research problems, strong classical and quantum baselines, realistic threat models, reproducible benchmarks, and shared best practices for secure and trustworthy QML systems in security-critical settings.\n\nKey Problems We Aim to Address\n- Trustworthy QML Systems: Define how quantum machine learning systems should be evaluated for security, robustness, privacy, reliability, interpretability, and model leakage. Contributions may study how QML models behave under noisy hardware, limited data, adversarial manipulation, and distribution shift.\n- Benchmarks and Evaluation: Develop realistic benchmarks, datasets, threat models, and evaluation protocols for secure and trustworthy QML. Contributions may compare QML methods against strong classical, quantum, and quantum-inspired baselines, including negative results and studies showing when QML is useful, limited, or unlikely to help.\n- Cyberdefense Applications: Explore QML and hybrid quantum-classical models for cyberdefense tasks such as intrusion detection, anomaly detection, malware analysis, phishing and fraud detection, vulnerability prioritization, cyber-physical security, and threat intelligence.\n- Secure Deployment: Study how QML pipelines can be securely trained, deployed, and accessed in cloud, edge, quantum-cloud, and hybrid computing environments. Contributions may address deployment risks, quantum-cloud access, secure quantum computing, quantum-network security, and post-quantum cyberdefense.\n\nCall for Papers\nNeurIPS 2026 Workshop on SaTQuML: Secure and Trustworthy Quantum Machine Learning invites submissions from researchers working on secure, reliable, and realistic quantum machine learning. We welcome work on trustworthy QML systems, rigorous benchmarking, cyberdefense applications, and secure deployment.\n\nKey Dates\n- Submission Start: August 15, 2026 (AoE)\n- Submission Deadline: August 29, 2026 (AoE)\n- Acceptance Notification: September 29, 2026 (AoE)\n- Camera-ready Deadline: October 15, 2026 (AoE)\n\nSubmission Site\nSubmit papers through the SaTQuML submission portal on OpenReview.\n\nSubmission Tracks\n- Long Paper: 9 pages.\n- Short Paper: 4 pages.\n- Tiny Paper: 2 pages.\n\nScope\nPrimary themes for the workshop include:\n- Trustworthy QML Systems: Adversarial robustness, privacy, reliability, model leakage, hardware noise, and distribution shift.\n- Rigorous Benchmarks and Evaluation: Realistic datasets and threat models; strong baselines; reproducibility, ablations, and negative results.\n- QML for Cyberdefense: Intrusion, anomaly, malware, fraud, and cyber-physical defense; vulnerability prioritization and threat intelligence.\n- Secure QML Deployment: Secure training and inference across cloud, edge, quantum-cloud, and hybrid systems; remote-access and pipeline risks.\n\nExample subtopics and concrete directions include:\n- Security and trustworthiness of QML systems, including robustness, privacy, fairness, reliability, interpretability, and model leakage.\n- Adversarial evaluation of QML, including perturbation, poisoning, evasion, extraction, quantum noise, and distribution-shift settings.\n- Security-sensitive learning methods, including quantum kernels, variational quantum circuits, quantum neural networks, quantum-inspired learning, and hybrid quantum-classical models.\n- Benchmarking and reproducibility, including realistic datasets, strong classical, quantum, and quantum-inspired baselines, ablations, and hardware-aware evaluation.\n- Cyberdefense applications such as intrusion detection, anomaly detection, malware analysis, phishing and fraud detection, cyber-physical security, and threat intelligence.\n- Practical utility and limitations of QML, including studies identifying when QML is useful, limited, or unlikely to improve over classical methods.\n- Secure deployment of QML pipelines in cloud, edge, quantum-cloud, and hybrid computing environments.\n- Quantum-era security, including post-quantum cyberdefense, quantum-network security, secure quantum computing, and trustworthy AI methods for quantum and hybrid models.\n\nSubmission Guidelines\nFormat: All submissions must be a single PDF file. We accept long papers up to 9 pages, short papers up to 4 pages, and tiny papers up to 2 pages. References and appendices are not included in the page limit, but the main text must be self-contained. Reviewers are not required to read beyond the main text.\nStyle file: You must format your submission using the NeurIPS 2026 LaTeX style file. Please include the references and supplementary materials in the same PDF. The maximum file size for submissions is 50MB. Submissions that violate the NeurIPS style (e.g., by decreasing margins or font sizes) or page limits may be rejected without further review.\nDual-submission and non-archival policy: We welcome ongoing and unpublished work. We will also accept papers that are under review at the time of submission, or that have been recently accepted, provided they do not breach any dual-submission or anonymity policies of those venues. The workshop is a non-archival venue and will not have official proceedings. Workshop submissions can be subsequently or concurrently submitted to other venues.\nVisibility: Submissions and reviews will not be public. Only accepted papers will be made public.\nDouble-blind reviewing: All submissions must be anonymized and may not contain any identifying information that may violate the double-blind reviewing policy. This policy applies to any supplementary or linked material as well, including code. If you are including links to any external material, it is your responsibility to guarantee anonymous browsing. Please do not include acknowledgements at submission time. If you need to cite one of your own papers, you should do so with adequate anonymization to preserve double-blind reviewing. Any papers found to be violating this policy will be rejected.\nPlease be AWARE: OpenReview's moderation policy for newly created profiles in the Call for Papers: New profiles created without an institutional email will go through a moderation process that can take up to two weeks. New profiles created with an institutional email will be activated automatically.\nContact: For any questions, please contact us at SaTQuML@gmail.com" }, { "key": "NeurReps", "title": "NeurIPS 2026 Workshop on Symmetry and Geometry in Neural Representations (Findings Track)", "subtitle": "NeurIPS 2026 Workshop NeurReps Findings", "summary": "NeurReps (Symmetry and Geometry in Neural Representations) gathers mathematicians, machine learning researchers, and neuroscientists to uncover the geometric and topological principles shared by brains and machines, spanning geometric/topological deep learning, computational and theoretical neuroscience, geometric statistics, and topological data analysis.", "cfp_full": "Symmetry and Geometry in Neural Representations\nNeurIPS 2026 Workshop - Sydney, Australia\nFri Dec 11 - Sat Dec 12, 2026 (exact day within this range to be confirmed by NeurIPS).\n\nA gathering of mathematicians, machine learning researchers, and neuroscientists working to uncover the geometric principles shared by brains and machines.\n\nThe fields of biological and artificial intelligence are converging on a shared principle: the geometry and topology of real-world structure play a central role in building efficient, robust, and interpretable representations.\n\nIn neuroscience, mounting evidence suggests that neural circuits encode task and environmental structure through low-dimensional manifolds, conserved symmetries, and structured transformations. In deep learning, sparsity, equivariance, and compositionality are guiding the development of more generalizable and interpretable models.\n\nNeurReps brings these threads together - fostering dialogue among machine learning researchers, neuroscientists, and mathematicians working to uncover unifying geometric principles of neural computation.\n\nJust as geometry and symmetry once unified the models of twentieth-century physics, we believe they will now illuminate the computational foundations of intelligence.\n\nCall for Papers\nThree tracks, one poster session - a subset of submissions will be selected for spotlight talks.\nSubmission Deadline: August 22, 2026 - AoE\nAccept / Reject Notification: September 29, 2026\n\nProceedings Track (9 pages, excl. refs + appendices)\nSelf-contained, highly-developed research papers. Archivally published in a dedicated PMLR volume. Double-blind review via OpenReview.\n\nExtended Abstract Track (4 pages, excl. refs + appendices)\nEarly-stage results, negative findings, opinion pieces, or novel datasets. Non-archival - may be posted to arXiv. Double-blind review via OpenReview.\n\nFindings Track (no page limit)\nNew this year: high-impact collaborative work between experimentalists and theorists, in any standard preprint format. Single-blind, editorially reviewed by an advisory panel of experts in the field.\n\nWhat We're Looking For\nWe invite submissions of novel research at the intersection of applied mathematics, deep learning, and computational neuroscience - work that incorporates symmetry and geometry into neural network design, the mechanistic interpretability of neural systems (biological or artificial), or theories of neural computation. We welcome contributions spanning geometric and topological deep learning, computational and theoretical neuroscience, geometric statistics, and topological data analysis.\n\nParticularly relevant themes\n- Theory and methods for learning invariant and equivariant representations\n- Statistical learning theory in the context of topology, geometry, and symmetry\n- Representational geometry in neural data\n- Learning and leveraging group structure in data\n- Equivariant world models for robotics\n- Dynamics of neural representations\n- Topological deep learning and topological data analysis\n- Geometric structure in language\n- Geometric and topological analysis of generative models\n- Symmetries, dynamical systems, and learning\n\nWe hope to see both theoretical contributions and applied results in domains including vision, motor control, navigation, and language, as well as the use of diverse mathematical objects such as quotient spaces, fiber bundles, Lie groups, Riemannian manifolds, graphs, topological domains, and group representations. We also welcome benchmark datasets and software. This list is guidance, not exhaustive - if you are unsure whether your work is in scope, please reach out to the organizers.\n\nA Novel Findings Track\nNeurReps has long been a primary home for broad computational and theoretical neuroscience work outside the scope of traditional machine learning venues. To honor and extend NeurIPS's historic ties to systems neuroscience, we are introducing a Findings Track for high-impact collaborative work between experimentalists and theorists - early versions of work of the caliber published in venues such as Cell, Nature, or Science, with the goal of early community exposure and dialogue between ML researchers and experimental neuroscientists.\nThe track is designed with minimal barriers to entry: no page limits, and any standard preprint format is welcome. No complex machine learning or deep learning is required - just some form of geometry, topology, or algebra. For example, work such as Gardner et al.'s Toroidal topology of population activity in grid cells (Nature, 2022) would be a natural fit. Because lab and dataset identity are often inseparable from the work, submissions are single-blind. Contributors present as posters, with a subset selected for spotlight talks.\n\nDual Submission Policy\nPapers in the Proceedings Track will be archivally published. Thus, submissions containing content that has been published or is under review elsewhere must include at least 30% new, unpublished/unsubmitted material. Likewise, to publish a NeurReps paper in another venue down the line, authors must add at least 30% new material. There are no restrictions on Extended Abstract submissions.\nAll submitting authors need an OpenReview profile for the Proceedings and Extended Abstract tracks - creating one can take a few days, so please don't wait until the deadline. For the Findings track, only the submitting author needs a profile; co-authors can be added by email.", "cfp_status": "published", "topics": [ "Theory and methods for learning invariant and equivariant representations", "Statistical learning theory in the context of topology, geometry, and symmetry", "Representational geometry in neural data", "Learning and leveraging group structure in data", "Equivariant world models for robotics", "Dynamics of neural representations", "Topological deep learning and topological data analysis", "Geometric structure in language", "Geometric and topological analysis of generative models", "Symmetries, dynamical systems, and learning" ], "important_dates": [ { "label": "Submission Deadline (all tracks, AoE)", "date": "2026-08-22" }, { "label": "Accept / Reject Notification", "date": "2026-09-29" }, { "label": "Workshop", "date": "2026-12-11" } ], "organizers": [], "speakers": [], "host_url": "https://neurreps.org/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/NeurReps_Proceedings", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/NeurReps_Findings", "location": "Sydney, New South Wales, Australia", "city": "Sydney", "workshop_date": "2026-12-11", "contact": "organizers@neurreps.org", "tracks": [ { "key": "NeurReps_Findings", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/NeurReps_Findings", "submission_dates_raw": "Submission Deadline: Aug 23 2026 11:59AM UTC-0" }, { "key": "NeurReps_Extended_Abstracts", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/NeurReps_Extended_Abstracts", "submission_dates_raw": "Submission Deadline: Aug 23 2026 11:59AM UTC-0" }, { "key": "NeurReps_Proceedings", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/NeurReps_Proceedings", "submission_dates_raw": "Submission Deadline: Aug 23 2026 11:59AM UTC-0" } ], "group": "theory", "group_label": "Theory & Methods", "corpus": "NeurIPS 2026 Workshop on Symmetry and Geometry in Neural Representations (Findings Track) NeurIPS 2026 Workshop NeurReps Findings NeurReps (Symmetry and Geometry in Neural Representations) gathers mathematicians, machine learning researchers, and neuroscientists to uncover the geometric and topological principles shared by brains and machines, spanning geometric/topological deep learning, computational and theoretical neuroscience, geometric statistics, and topological data analysis. Theory and methods for learning invariant and equivariant representations Statistical learning theory in the context of topology, geometry, and symmetry Representational geometry in neural data Learning and leveraging group structure in data Equivariant world models for robotics Dynamics of neural representations Topological deep learning and topological data analysis Geometric structure in language Geometric and topological analysis of generative models Symmetries, dynamical systems, and learning Symmetry and Geometry in Neural Representations\nNeurIPS 2026 Workshop - Sydney, Australia\nFri Dec 11 - Sat Dec 12, 2026 (exact day within this range to be confirmed by NeurIPS).\n\nA gathering of mathematicians, machine learning researchers, and neuroscientists working to uncover the geometric principles shared by brains and machines.\n\nThe fields of biological and artificial intelligence are converging on a shared principle: the geometry and topology of real-world structure play a central role in building efficient, robust, and interpretable representations.\n\nIn neuroscience, mounting evidence suggests that neural circuits encode task and environmental structure through low-dimensional manifolds, conserved symmetries, and structured transformations. In deep learning, sparsity, equivariance, and compositionality are guiding the development of more generalizable and interpretable models.\n\nNeurReps brings these threads together - fostering dialogue among machine learning researchers, neuroscientists, and mathematicians working to uncover unifying geometric principles of neural computation.\n\nJust as geometry and symmetry once unified the models of twentieth-century physics, we believe they will now illuminate the computational foundations of intelligence.\n\nCall for Papers\nThree tracks, one poster session - a subset of submissions will be selected for spotlight talks.\nSubmission Deadline: August 22, 2026 - AoE\nAccept / Reject Notification: September 29, 2026\n\nProceedings Track (9 pages, excl. refs + appendices)\nSelf-contained, highly-developed research papers. Archivally published in a dedicated PMLR volume. Double-blind review via OpenReview.\n\nExtended Abstract Track (4 pages, excl. refs + appendices)\nEarly-stage results, negative findings, opinion pieces, or novel datasets. Non-archival - may be posted to arXiv. Double-blind review via OpenReview.\n\nFindings Track (no page limit)\nNew this year: high-impact collaborative work between experimentalists and theorists, in any standard preprint format. Single-blind, editorially reviewed by an advisory panel of experts in the field.\n\nWhat We're Looking For\nWe invite submissions of novel research at the intersection of applied mathematics, deep learning, and computational neuroscience - work that incorporates symmetry and geometry into neural network design, the mechanistic interpretability of neural systems (biological or artificial), or theories of neural computation. We welcome contributions spanning geometric and topological deep learning, computational and theoretical neuroscience, geometric statistics, and topological data analysis.\n\nParticularly relevant themes\n- Theory and methods for learning invariant and equivariant representations\n- Statistical learning theory in the context of topology, geometry, and symmetry\n- Representational geometry in neural data\n- Learning and leveraging group structure in data\n- Equivariant world models for robotics\n- Dynamics of neural representations\n- Topological deep learning and topological data analysis\n- Geometric structure in language\n- Geometric and topological analysis of generative models\n- Symmetries, dynamical systems, and learning\n\nWe hope to see both theoretical contributions and applied results in domains including vision, motor control, navigation, and language, as well as the use of diverse mathematical objects such as quotient spaces, fiber bundles, Lie groups, Riemannian manifolds, graphs, topological domains, and group representations. We also welcome benchmark datasets and software. This list is guidance, not exhaustive - if you are unsure whether your work is in scope, please reach out to the organizers.\n\nA Novel Findings Track\nNeurReps has long been a primary home for broad computational and theoretical neuroscience work outside the scope of traditional machine learning venues. To honor and extend NeurIPS's historic ties to systems neuroscience, we are introducing a Findings Track for high-impact collaborative work between experimentalists and theorists - early versions of work of the caliber published in venues such as Cell, Nature, or Science, with the goal of early community exposure and dialogue between ML researchers and experimental neuroscientists.\nThe track is designed with minimal barriers to entry: no page limits, and any standard preprint format is welcome. No complex machine learning or deep learning is required - just some form of geometry, topology, or algebra. For example, work such as Gardner et al.'s Toroidal topology of population activity in grid cells (Nature, 2022) would be a natural fit. Because lab and dataset identity are often inseparable from the work, submissions are single-blind. Contributors present as posters, with a subset selected for spotlight talks.\n\nDual Submission Policy\nPapers in the Proceedings Track will be archivally published. Thus, submissions containing content that has been published or is under review elsewhere must include at least 30% new, unpublished/unsubmitted material. Likewise, to publish a NeurReps paper in another venue down the line, authors must add at least 30% new material. There are no restrictions on Extended Abstract submissions.\nAll submitting authors need an OpenReview profile for the Proceedings and Extended Abstract tracks - creating one can take a few days, so please don't wait until the deadline. For the Findings track, only the submitting author needs a profile; co-authors can be added by email." }, { "key": "TTCL", "title": "NeurIPS 2026 Workshop on Towards Test-Time Continual Learning Agents", "subtitle": "NeurIPS 2026 TTCL Workshop", "summary": "A NeurIPS 2026 workshop on agents that keep learning after deployment, uniting test-time, continual, and agentic learning so that AI systems continuously acquire, consolidate, and refine knowledge and skills at inference time without catastrophic forgetting.", "cfp_full": "About the Workshop\n\nThe rapid progress of foundation models has produced agents with remarkable capabilities in perception, language understanding, reasoning, and tool use, while advances in post-training, reinforcement learning, retrieval-augmented generation, and agentic scaffolding now let both virtual and embodied agents tackle complex tasks in coding, web navigation, scientific discovery, and robotic control. Yet today's agents remain largely static: after costly pre- and post-training, their knowledge, skills, and behaviors are mostly fixed, and once deployed they adapt through prompting, retrieval, or external tools rather than genuine internal learning and memory consolidation.\n\nThis contrasts sharply with human intelligence, where people continuously acquire knowledge, refine representations, and reorganize beliefs through interaction. Current agents, whether virtual or embodied, instead cannot continually learn at test time: they struggle to internalize new information after deployment, fail to improve from repeated mistakes, and can forget prior knowledge and skills when updated naively (catastrophic forgetting).\n\nThe NeurIPS 2026 Workshop on Towards Test-Time Continual Learning Agents (TTCL) brings together researchers across continual learning, large language models, reinforcement learning, embodied AI, memory systems, cognitive science, robotics, and multimodal learning. We define Test-Time Continual Learning Agents as AI systems that continuously acquire, consolidate, and refine knowledge and capabilities during deployment, without catastrophic forgetting or repeated large-scale retraining. This goes beyond updating facts: it asks how agents improve perception, reasoning, planning, exploration, skill acquisition, and long-term decision-making through ongoing experience in virtual and physical worlds. These capabilities are especially important for robotics, scientific discovery, personalized assistants, education, healthcare, and human-AI collaboration.\n\nOur central vision is to catalyze progress toward next-generation cognitive agents that learn continually at test time, consolidate experience over long horizons, improve through interaction, and remain reliable in dynamic physical and virtual worlds, rethinking the boundaries between training and inference, memory and learning, adaptation and reasoning.\n\nCall for Papers\n\nWe invite contributions at the intersection of test-time learning, continual learning, and agentic AI. Relevant topics include (but are not limited to):\n\nTopics\n- Test-time adaptation and learning. Test-time training and adaptation, online and meta-learning, in-context and parameter-efficient fine-tuning, and self-improvement from environment feedback and self-generated supervision.\n- Continual learning without catastrophic forgetting. Regularization-, replay-, optimization-, representation-, and architecture-based approaches to the stability-plasticity trade-off, applied to models and agents that must accumulate knowledge and skills across tasks.\n- Memory and knowledge consolidation. Episodic and semantic memory architectures, retrieval-augmented generation, and consolidation mechanisms that transform interaction trajectories into persistent knowledge and reusable skills.\n- Agents that learn from experience. Language, multimodal, and embodied agents that improve exploration, long-horizon planning, and skill acquisition from experience, including intrinsic motivation and open-ended learning.\n- Evaluation, safety, and robustness. Benchmarks and evaluation protocols for long-horizon test-time continual learning, and the safety, robustness, and alignment of self-improving agents.\n\nSubmissions are managed via OpenReview; see the Submission Guidelines below.\n\nSubmission Tracks\n- General Research Track (4-9 pages): Novel frameworks, empirical studies, algorithmic advances, benchmarks, position papers, or system demonstrations at the test-time x continual x agentic intersection.\n- Challenge Track (4-9 pages): Papers accompanying entries to the AgentOdyssey Challenge: either a general research paper that includes AgentOdyssey benchmark results, or a technical report describing an agent implemented for AgentOdyssey evaluation. See the Challenge section for participation rules.\n\nAwards\n\nThanks to the generous sponsorship of Lambda, the workshop will present:\n- Best Paper Award: $3,000 in compute credits, selected from all submitted papers across both tracks (General Research and Challenge)\n- Two Runner-up Awards: $1,500 in compute credits each\n- Every accepted paper: $400 in compute credits\n- Challenge Award: prize TBD, open exclusively to Challenge Track submissions and announced at the workshop; see the Challenge section for eligibility and ranking rules\n\nAwards recognize the strongest contributions to the workshop.\n\nSubmission Guidelines\n\nFormatting. Submissions must be in English, follow the NeurIPS 2026 LaTeX template, and be submitted as a single PDF via OpenReview. The page limits above apply to the main text; references and appendices are not included in the page limit, but the main text must be self-contained, and reviewers are not required to read beyond it. Submissions exceeding the page limit will be desk-rejected.\n\nAnonymity. The workshop follows a double-blind review process. Submissions must be anonymized by removing author names, affiliations, and acknowledgments. Prior work should be cited in the third person, and identifying information, including in supplementary materials, must be omitted. Reviewing follows the NeurIPS conflict-of-interest guidelines, and organizers will not submit papers to TTCL.\n\nDual Submission & Non-Archival Policy. The workshop is non-archival: we welcome submissions that are under review at, or have been accepted by, other venues. Papers already accepted to the NeurIPS 2026 main conference will undergo an expedited review process primarily evaluating their relevance to the workshop themes. Accepted papers will be made publicly available on OpenReview, and all accepted papers will be presented in a poster session.\n\nImportant Dates\n- Submission deadline: August 29, 2026\n- Notification of acceptance: September 25, 2026\n- Camera-ready due: October 25, 2026\n- Workshop day (exact day TBA): December 12 or 13, 2026\nAll deadlines are 11:59 PM, Anywhere on Earth (AoE).", "cfp_status": "published", "topics": [ "Test-time adaptation and learning (test-time training, online and meta-learning, in-context and parameter-efficient fine-tuning, self-improvement from feedback)", "Continual learning without catastrophic forgetting (regularization-, replay-, optimization-, representation-, and architecture-based approaches to stability-plasticity)", "Memory and knowledge consolidation (episodic and semantic memory, retrieval-augmented generation, consolidation of interaction trajectories into skills)", "Agents that learn from experience (language, multimodal, and embodied agents improving exploration, long-horizon planning, skill acquisition, open-ended learning)", "Evaluation, safety, and robustness (benchmarks and protocols for long-horizon test-time continual learning; safety, robustness, alignment of self-improving agents)" ], "important_dates": [ { "label": "Submission deadline", "date": "2026-08-29" }, { "label": "Notification of acceptance", "date": "2026-09-25" }, { "label": "Camera-ready due", "date": "2026-10-25" }, { "label": "Workshop day (exact day TBA)", "date": "December 12 or 13, 2026" } ], "organizers": [ "Zheyuan \"Brian\" Zhang (Johns Hopkins University)", "Chuanyang Jin (Johns Hopkins University)", "Jacob Sansom (University of Michigan)", "Zekun Wang (Georgia Institute of Technology)", "Jianwen Xie (Lambda)", "Joyce Chai (University of Michigan)", "Daniel Khashabi (Johns Hopkins University)", "Tianmin Shu (Johns Hopkins University)" ], "speakers": [ "Christopher MacLellan (Georgia Tech)", "Manling Li (Northwestern University)", "Zsolt Kira (Georgia Tech)", "Yu Su (Ohio State University & NeoCognition)", "Kelsey Allen (University of British Columbia)", "Cansu Sancaktar (Max Planck Institute for Intelligent Systems)", "Sebastian Risi (IT University of Copenhagen)", "Jiafei Duan (University of Washington, incoming NUS)", "Jason Weston (Meta, TBC)", "Chelsea Finn (Stanford University & Physical Intelligence, TBC)" ], "host_url": "https://ttcl-agents.github.io", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/TTCL", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/TTCL", "location": "Atlanta, United States", "city": "Atlanta", "workshop_date": "2026-07-20", "contact": "zzhan378@jhu.edu", "tracks": [ { "key": "TTCL", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/TTCL", "submission_dates_raw": "" } ], "group": "theory", "group_label": "Theory & Methods", "corpus": "NeurIPS 2026 Workshop on Towards Test-Time Continual Learning Agents NeurIPS 2026 TTCL Workshop A NeurIPS 2026 workshop on agents that keep learning after deployment, uniting test-time, continual, and agentic learning so that AI systems continuously acquire, consolidate, and refine knowledge and skills at inference time without catastrophic forgetting. Test-time adaptation and learning (test-time training, online and meta-learning, in-context and parameter-efficient fine-tuning, self-improvement from feedback) Continual learning without catastrophic forgetting (regularization-, replay-, optimization-, representation-, and architecture-based approaches to stability-plasticity) Memory and knowledge consolidation (episodic and semantic memory, retrieval-augmented generation, consolidation of interaction trajectories into skills) Agents that learn from experience (language, multimodal, and embodied agents improving exploration, long-horizon planning, skill acquisition, open-ended learning) Evaluation, safety, and robustness (benchmarks and protocols for long-horizon test-time continual learning; safety, robustness, alignment of self-improving agents) About the Workshop\n\nThe rapid progress of foundation models has produced agents with remarkable capabilities in perception, language understanding, reasoning, and tool use, while advances in post-training, reinforcement learning, retrieval-augmented generation, and agentic scaffolding now let both virtual and embodied agents tackle complex tasks in coding, web navigation, scientific discovery, and robotic control. Yet today's agents remain largely static: after costly pre- and post-training, their knowledge, skills, and behaviors are mostly fixed, and once deployed they adapt through prompting, retrieval, or external tools rather than genuine internal learning and memory consolidation.\n\nThis contrasts sharply with human intelligence, where people continuously acquire knowledge, refine representations, and reorganize beliefs through interaction. Current agents, whether virtual or embodied, instead cannot continually learn at test time: they struggle to internalize new information after deployment, fail to improve from repeated mistakes, and can forget prior knowledge and skills when updated naively (catastrophic forgetting).\n\nThe NeurIPS 2026 Workshop on Towards Test-Time Continual Learning Agents (TTCL) brings together researchers across continual learning, large language models, reinforcement learning, embodied AI, memory systems, cognitive science, robotics, and multimodal learning. We define Test-Time Continual Learning Agents as AI systems that continuously acquire, consolidate, and refine knowledge and capabilities during deployment, without catastrophic forgetting or repeated large-scale retraining. This goes beyond updating facts: it asks how agents improve perception, reasoning, planning, exploration, skill acquisition, and long-term decision-making through ongoing experience in virtual and physical worlds. These capabilities are especially important for robotics, scientific discovery, personalized assistants, education, healthcare, and human-AI collaboration.\n\nOur central vision is to catalyze progress toward next-generation cognitive agents that learn continually at test time, consolidate experience over long horizons, improve through interaction, and remain reliable in dynamic physical and virtual worlds, rethinking the boundaries between training and inference, memory and learning, adaptation and reasoning.\n\nCall for Papers\n\nWe invite contributions at the intersection of test-time learning, continual learning, and agentic AI. Relevant topics include (but are not limited to):\n\nTopics\n- Test-time adaptation and learning. Test-time training and adaptation, online and meta-learning, in-context and parameter-efficient fine-tuning, and self-improvement from environment feedback and self-generated supervision.\n- Continual learning without catastrophic forgetting. Regularization-, replay-, optimization-, representation-, and architecture-based approaches to the stability-plasticity trade-off, applied to models and agents that must accumulate knowledge and skills across tasks.\n- Memory and knowledge consolidation. Episodic and semantic memory architectures, retrieval-augmented generation, and consolidation mechanisms that transform interaction trajectories into persistent knowledge and reusable skills.\n- Agents that learn from experience. Language, multimodal, and embodied agents that improve exploration, long-horizon planning, and skill acquisition from experience, including intrinsic motivation and open-ended learning.\n- Evaluation, safety, and robustness. Benchmarks and evaluation protocols for long-horizon test-time continual learning, and the safety, robustness, and alignment of self-improving agents.\n\nSubmissions are managed via OpenReview; see the Submission Guidelines below.\n\nSubmission Tracks\n- General Research Track (4-9 pages): Novel frameworks, empirical studies, algorithmic advances, benchmarks, position papers, or system demonstrations at the test-time x continual x agentic intersection.\n- Challenge Track (4-9 pages): Papers accompanying entries to the AgentOdyssey Challenge: either a general research paper that includes AgentOdyssey benchmark results, or a technical report describing an agent implemented for AgentOdyssey evaluation. See the Challenge section for participation rules.\n\nAwards\n\nThanks to the generous sponsorship of Lambda, the workshop will present:\n- Best Paper Award: $3,000 in compute credits, selected from all submitted papers across both tracks (General Research and Challenge)\n- Two Runner-up Awards: $1,500 in compute credits each\n- Every accepted paper: $400 in compute credits\n- Challenge Award: prize TBD, open exclusively to Challenge Track submissions and announced at the workshop; see the Challenge section for eligibility and ranking rules\n\nAwards recognize the strongest contributions to the workshop.\n\nSubmission Guidelines\n\nFormatting. Submissions must be in English, follow the NeurIPS 2026 LaTeX template, and be submitted as a single PDF via OpenReview. The page limits above apply to the main text; references and appendices are not included in the page limit, but the main text must be self-contained, and reviewers are not required to read beyond it. Submissions exceeding the page limit will be desk-rejected.\n\nAnonymity. The workshop follows a double-blind review process. Submissions must be anonymized by removing author names, affiliations, and acknowledgments. Prior work should be cited in the third person, and identifying information, including in supplementary materials, must be omitted. Reviewing follows the NeurIPS conflict-of-interest guidelines, and organizers will not submit papers to TTCL.\n\nDual Submission & Non-Archival Policy. The workshop is non-archival: we welcome submissions that are under review at, or have been accepted by, other venues. Papers already accepted to the NeurIPS 2026 main conference will undergo an expedited review process primarily evaluating their relevance to the workshop themes. Accepted papers will be made publicly available on OpenReview, and all accepted papers will be presented in a poster session.\n\nImportant Dates\n- Submission deadline: August 29, 2026\n- Notification of acceptance: September 25, 2026\n- Camera-ready due: October 25, 2026\n- Workshop day (exact day TBA): December 12 or 13, 2026\nAll deadlines are 11:59 PM, Anywhere on Earth (AoE)." }, { "key": "RTCA", "title": "NeurIPS 2026 Workshop Real-Time Conversational Agents: Toward Natural Multimodal Interaction", "subtitle": "NeurIPS 2026 Workshop RTCA", "summary": "RTCA explores the challenges of real-time conversational agents across multiple modalities, focusing on latency, turn-taking, cross-modal alignment, and the evaluation of interactional naturalness in live multimodal systems.", "cfp_full": "Real-Time Conversational Agents: Toward Natural Multimodal Interaction\nRTCA looks to explore the challenges of realtime conversational agents across multiple modalities.\n\nWhy RTCA\nReal-time interaction is a different research problem.\nConversational AI has moved beyond text chat into voice modes, visual avatars, shared screens, and tools. Systems that feel natural must stream speech, video, and language while listening, watching, and re-planning continuously with challenging latency constraints.\nRTCA focuses on the problems that offline generation can afford to ignore: latency, partial observability, turn-taking, backchannels, interruptions, cross-modal alignment, and evaluation of interactional naturalness.\n\n01 Real-Time Generation: Streaming speech, video, and language under hard latency budgets.\n02 Naturalness in Interaction: Prosody, gaze, timing, grounding, expressivity, and turn-taking dynamics.\n03 Evaluation of Live Systems: Metrics and protocols for responsiveness, perceived latency, and conversational quality.\n\nScope\nTopics of interest\nWe invite original contributions from speech, vision, language, HCI, social-signal processing, and ML systems communities working on interactive multimodal agents, including but not limited to:\n- Streaming/low-latency speech synthesis, ASR, and full-duplex audio-language models\n- Real-time talking-head, avatar, and embodied video generation; lip-sync, gaze, expressivity under streaming\n- Streaming language models; incremental and speculative decoding for dialogue\n- Turn-taking, backchanneling, interruption handling, and floor management\n- Multimodal alignment under latency and partial-observation constraints\n- Prosody, emotion, and paralinguistic generation in interactive settings\n- Memory, grounding, and tool use during live conversation\n- Evaluation of naturalness: perceptual studies, turn-taking metrics, perceived latency, interactive Turing-style tests\n- Datasets and benchmarks for interactive (not offline) evaluation\n- Efficient inference, on-device deployment, and the systems-quality trade-off\n- Safety, identity, and trust in real-time agents (deepfakes, persuasion, consent)\n\nSubmissions\nCall for papers\nThe call for papers opens July 18, 2026. Submissions (papers and demos) are due August 29, 2026 AoE via OpenReview, formatted for double-blind review using the NeurIPS 2026 style file.\n\nFull papers: Up to 8 pages. Original contributions; may be presented as posters and/or contributed talks.\nShort papers: Up to 4 pages. Work in progress or focused contributions.\nDemo papers: Up to 2 pages.\n\nImportant dates\nJuly 18 - Call for papers opens\nAug. 29 AoE - Submission deadline (papers and demos)\nSep. 29 AoE - Author notification\nDec. 11 or 12 - Workshop day, Sydney, Australia\n\nPage limits exclude references and appendices. The workshop is non-archival; authors retain the right to publish elsewhere. Dataset submissions should include an ethics statement covering consent for voice or likeness, deepfake risk, and provenance. The workshop follows the NeurIPS Code of Ethics, Code of Conduct, and Main Track LLM policy. Questions? Contact us at rtca-workshop@googlegroups.com.", "cfp_status": "published", "topics": [ "Streaming/low-latency speech synthesis, ASR, and full-duplex audio-language models", "Real-time talking-head, avatar, and embodied video generation; lip-sync, gaze, expressivity under streaming", "Streaming language models; incremental and speculative decoding for dialogue", "Turn-taking, backchanneling, interruption handling, and floor management", "Multimodal alignment under latency and partial-observation constraints", "Prosody, emotion, and paralinguistic generation in interactive settings", "Memory, grounding, and tool use during live conversation", "Evaluation of naturalness: perceptual studies, turn-taking metrics, perceived latency, interactive Turing-style tests", "Datasets and benchmarks for interactive (not offline) evaluation", "Efficient inference, on-device deployment, and the systems-quality trade-off", "Safety, identity, and trust in real-time agents (deepfakes, persuasion, consent)" ], "important_dates": [ { "label": "Call for papers opens", "date": "2026-07-18" }, { "label": "Submission deadline (papers and demos)", "date": "2026-08-29" }, { "label": "Author notification", "date": "2026-09-29" }, { "label": "Workshop day", "date": "Dec. 11 or 12" } ], "organizers": [ "Niki Foteinopoulou (Research Scientist, Tavus)", "Alessandro Conti (Research Scientist, Tavus)", "Jack Saunders (Senior Research Scientist, Tavus)", "Oya Celiktutan (Reader in AI and Robotics, King's College London)", "Cigdem Beyan (Associate Professor, University of Verona)", "Ioannis Patras (Professor, Queen Mary University of London)" ], "speakers": [], "host_url": "https://rtcaneurips26.github.io/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/RTCA", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/RTCA", "location": "Sydney, Australia", "city": "Sydney", "workshop_date": "2026-12-10", "contact": "rtca-workshop@googlegroups.com", "tracks": [ { "key": "RTCA", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/RTCA", "submission_dates_raw": "" } ], "group": "agents", "group_label": "Agents & Agentic Systems", "corpus": "NeurIPS 2026 Workshop Real-Time Conversational Agents: Toward Natural Multimodal Interaction NeurIPS 2026 Workshop RTCA RTCA explores the challenges of real-time conversational agents across multiple modalities, focusing on latency, turn-taking, cross-modal alignment, and the evaluation of interactional naturalness in live multimodal systems. Streaming/low-latency speech synthesis, ASR, and full-duplex audio-language models Real-time talking-head, avatar, and embodied video generation; lip-sync, gaze, expressivity under streaming Streaming language models; incremental and speculative decoding for dialogue Turn-taking, backchanneling, interruption handling, and floor management Multimodal alignment under latency and partial-observation constraints Prosody, emotion, and paralinguistic generation in interactive settings Memory, grounding, and tool use during live conversation Evaluation of naturalness: perceptual studies, turn-taking metrics, perceived latency, interactive Turing-style tests Datasets and benchmarks for interactive (not offline) evaluation Efficient inference, on-device deployment, and the systems-quality trade-off Safety, identity, and trust in real-time agents (deepfakes, persuasion, consent) Real-Time Conversational Agents: Toward Natural Multimodal Interaction\nRTCA looks to explore the challenges of realtime conversational agents across multiple modalities.\n\nWhy RTCA\nReal-time interaction is a different research problem.\nConversational AI has moved beyond text chat into voice modes, visual avatars, shared screens, and tools. Systems that feel natural must stream speech, video, and language while listening, watching, and re-planning continuously with challenging latency constraints.\nRTCA focuses on the problems that offline generation can afford to ignore: latency, partial observability, turn-taking, backchannels, interruptions, cross-modal alignment, and evaluation of interactional naturalness.\n\n01 Real-Time Generation: Streaming speech, video, and language under hard latency budgets.\n02 Naturalness in Interaction: Prosody, gaze, timing, grounding, expressivity, and turn-taking dynamics.\n03 Evaluation of Live Systems: Metrics and protocols for responsiveness, perceived latency, and conversational quality.\n\nScope\nTopics of interest\nWe invite original contributions from speech, vision, language, HCI, social-signal processing, and ML systems communities working on interactive multimodal agents, including but not limited to:\n- Streaming/low-latency speech synthesis, ASR, and full-duplex audio-language models\n- Real-time talking-head, avatar, and embodied video generation; lip-sync, gaze, expressivity under streaming\n- Streaming language models; incremental and speculative decoding for dialogue\n- Turn-taking, backchanneling, interruption handling, and floor management\n- Multimodal alignment under latency and partial-observation constraints\n- Prosody, emotion, and paralinguistic generation in interactive settings\n- Memory, grounding, and tool use during live conversation\n- Evaluation of naturalness: perceptual studies, turn-taking metrics, perceived latency, interactive Turing-style tests\n- Datasets and benchmarks for interactive (not offline) evaluation\n- Efficient inference, on-device deployment, and the systems-quality trade-off\n- Safety, identity, and trust in real-time agents (deepfakes, persuasion, consent)\n\nSubmissions\nCall for papers\nThe call for papers opens July 18, 2026. Submissions (papers and demos) are due August 29, 2026 AoE via OpenReview, formatted for double-blind review using the NeurIPS 2026 style file.\n\nFull papers: Up to 8 pages. Original contributions; may be presented as posters and/or contributed talks.\nShort papers: Up to 4 pages. Work in progress or focused contributions.\nDemo papers: Up to 2 pages.\n\nImportant dates\nJuly 18 - Call for papers opens\nAug. 29 AoE - Submission deadline (papers and demos)\nSep. 29 AoE - Author notification\nDec. 11 or 12 - Workshop day, Sydney, Australia\n\nPage limits exclude references and appendices. The workshop is non-archival; authors retain the right to publish elsewhere. Dataset submissions should include an ethics statement covering consent for voice or likeness, deepfake risk, and provenance. The workshop follows the NeurIPS Code of Ethics, Code of Conduct, and Main Track LLM policy. Questions? Contact us at rtca-workshop@googlegroups.com." }, { "key": "Verify-Agents", "title": "NeurIPS 2026 Workshop Who Verifies the Agents? Toward Reliable Agent Development", "subtitle": "NeurIPS 2026 verify-agents", "summary": "This workshop treats verification as a first-class research problem for agent development, bringing together researchers and practitioners working on robust verifiers, environment-grounded evaluation, and richer verification signals to lay the foundations of reliable agent systems.", "cfp_full": "Who Verifies the Agents? Toward Reliable Agent Development\nVerification is the bottleneck between fragile prototypes and scalable, reliable agent systems. This workshop convenes researchers and practitioners to make verification a first-class discipline.\n\nOverview\nAgents that reason, plan, and act in open-ended environments are advancing at a remarkable pace. Yet a basic question has become surprisingly hard to answer: when we update an agent's prompt, add a new tool, or change its reasoning strategy, did it actually get better? Answering that question is verification. Today it works well only where ground truth is clear, such as formal mathematics, competitive programming, and software tests. For general agentic tasks, verification signals remain shallow and noisy: improvements plateau, regressions slip through silently, and development turns into guesswork.\nThis workshop treats verification as a first-class research problem. We bring together researchers and practitioners working on robust verifiers, environment-grounded evaluation, and richer verification signals to lay the foundations of reliable agent development.\n\nCall for Papers\nTopics of Interest\nWe invite submissions across three core pillars, as well as topics at their intersection:\n\nPillar 1: Safety and Robustness of Verification\n- Robust verifiers that prevent reward hacking and specification gaming\n- Adversarial robustness of verifiers and red-teaming of evaluation harnesses\n- Alignment-aware verification: ensuring verifiers remain faithful as agents evolve\n\nPillar 2: Environment-Grounded Verification and Simulators\n- Faithful simulators as verification infrastructure for open-ended tasks\n- Multi-agent and self-optimizing systems for environment-grounded evaluation\n- Measuring agents in production: observability, monitoring, and runtime verification\n- Evolutionary and search-based methods for environment-driven agent optimization\n- Benchmarks and environment design that stress-test verification methods\n\nPillar 3: Diverse and Heterogeneous Verifiable Signals\n- Composing heterogeneous signals (user experience, cost/latency, calibration, multimodality) into reliable verification metrics\n- Beyond scalar rewards: holistic evaluation of agentic behavior\n- Human-in-the-loop verification and human-AI collaborative evaluation\n- Reflective and self-improving verification (agents that verify other agents)\n- Automated agent design, prompt optimization, and scaffold search with verifiable feedback\n- Formal verification of agent-generated artifacts, including the use of proof assistants and verification languages (e.g., Dafny, Rocq, and Lean)\n\nCross-Cutting Topics\n- Verification for meta-agents and Agent4Agent systems (agents that design, optimize, or evaluate other agents)\n- Self-evolving agents: stable improvement without collapse or reward hacking\n- Evaluation of agent-generated designs vs. human-engineered systems\n- Scalable oversight and verification for long-horizon, multi-step agent behavior\n- Cognitive and neuroscience-inspired verification frameworks\n\nSubmission Guidelines\nFormat: Papers should be between 4 to 9 pages (excluding references and appendices), using the NeurIPS 2026 template. We also welcome demo papers, which should be no more than 4 pages.\nDual submission policy: We welcome work that is under review or has been recently published at other venues.\nReview: Reviews will be double blind. Authors of submitted papers may be asked to contribute reviews.\nPresentation: Accepted papers will be presented as posters; select papers will be chosen for oral presentations or lightning talks.\nNon-archival: The workshop is non-archival; accepted papers will be made available on OpenReview but do not constitute formal proceedings.\nSubmissions and reviewing are handled through OpenReview.\n\nImportant Dates\nSubmission deadline: August 29, 2026\nAuthor notification: September 29, 2026\nWorkshop day: Dec 11 or 12, 2026\nThe exact workshop day (Friday, Dec 11 or Saturday, Dec 12) will be confirmed once assigned by NeurIPS. All deadlines are 23:59 Anywhere on Earth (AoE) unless otherwise noted.", "cfp_status": "published", "topics": [ "Robust verifiers that prevent reward hacking and specification gaming", "Adversarial robustness of verifiers and red-teaming of evaluation harnesses", "Alignment-aware verification", "Faithful simulators as verification infrastructure for open-ended tasks", "Multi-agent and self-optimizing systems for environment-grounded evaluation", "Observability, monitoring, and runtime verification of agents in production", "Evolutionary and search-based methods for environment-driven agent optimization", "Benchmarks and environment design that stress-test verification methods", "Composing heterogeneous verifiable signals into reliable verification metrics", "Beyond scalar rewards: holistic evaluation of agentic behavior", "Human-in-the-loop and human-AI collaborative verification", "Reflective and self-improving verification (agents that verify other agents)", "Automated agent design, prompt optimization, and scaffold search with verifiable feedback", "Formal verification of agent-generated artifacts (Dafny, Rocq, Lean)", "Verification for meta-agents and Agent4Agent systems", "Self-evolving agents without collapse or reward hacking", "Scalable oversight for long-horizon, multi-step agent behavior", "Cognitive and neuroscience-inspired verification frameworks" ], "important_dates": [ { "label": "Submission deadline", "date": "2026-08-29" }, { "label": "Author notification", "date": "2026-09-29" }, { "label": "Workshop day", "date": "Dec 11 or 12, 2026" } ], "organizers": [ "Ahmad Beirami (Fidian)", "Mert Cemri (UC Berkeley)", "Zhang-Wei Hong (MIT-IBM Watson AI Lab)", "Hung Le (Fidian)", "Ninareh Mehrabi (Meta Superintelligence Labs)", "Melissa Pan (UC Berkeley)", "Dilara Soylu (Stanford University)" ], "speakers": [ "Pin-Yu Chen (IBM)", "Azalia Mirhoseini (Stanford · Ricursive Intelligence)", "Seshendra Nalla (Datadog)", "Dhaval Patel (IBM Research)", "Ion Stoica (UC Berkeley · Anyscale · Databricks)", "Yu Su (Ohio State · NeoCognition)" ], "host_url": "https://verify-agents-workshop.github.io/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/Verify-Agents", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/Verify-Agents", "location": "Sydney, Australia", "city": "Sydney", "workshop_date": "2026-12-10", "contact": "verify-agents-workshop@googlegroups.com", "tracks": [ { "key": "Verify-Agents", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/Verify-Agents", "submission_dates_raw": "" } ], "group": "agents", "group_label": "Agents & Agentic Systems", "corpus": "NeurIPS 2026 Workshop Who Verifies the Agents? Toward Reliable Agent Development NeurIPS 2026 verify-agents This workshop treats verification as a first-class research problem for agent development, bringing together researchers and practitioners working on robust verifiers, environment-grounded evaluation, and richer verification signals to lay the foundations of reliable agent systems. Robust verifiers that prevent reward hacking and specification gaming Adversarial robustness of verifiers and red-teaming of evaluation harnesses Alignment-aware verification Faithful simulators as verification infrastructure for open-ended tasks Multi-agent and self-optimizing systems for environment-grounded evaluation Observability, monitoring, and runtime verification of agents in production Evolutionary and search-based methods for environment-driven agent optimization Benchmarks and environment design that stress-test verification methods Composing heterogeneous verifiable signals into reliable verification metrics Beyond scalar rewards: holistic evaluation of agentic behavior Human-in-the-loop and human-AI collaborative verification Reflective and self-improving verification (agents that verify other agents) Automated agent design, prompt optimization, and scaffold search with verifiable feedback Formal verification of agent-generated artifacts (Dafny, Rocq, Lean) Verification for meta-agents and Agent4Agent systems Self-evolving agents without collapse or reward hacking Scalable oversight for long-horizon, multi-step agent behavior Cognitive and neuroscience-inspired verification frameworks Who Verifies the Agents? Toward Reliable Agent Development\nVerification is the bottleneck between fragile prototypes and scalable, reliable agent systems. This workshop convenes researchers and practitioners to make verification a first-class discipline.\n\nOverview\nAgents that reason, plan, and act in open-ended environments are advancing at a remarkable pace. Yet a basic question has become surprisingly hard to answer: when we update an agent's prompt, add a new tool, or change its reasoning strategy, did it actually get better? Answering that question is verification. Today it works well only where ground truth is clear, such as formal mathematics, competitive programming, and software tests. For general agentic tasks, verification signals remain shallow and noisy: improvements plateau, regressions slip through silently, and development turns into guesswork.\nThis workshop treats verification as a first-class research problem. We bring together researchers and practitioners working on robust verifiers, environment-grounded evaluation, and richer verification signals to lay the foundations of reliable agent development.\n\nCall for Papers\nTopics of Interest\nWe invite submissions across three core pillars, as well as topics at their intersection:\n\nPillar 1: Safety and Robustness of Verification\n- Robust verifiers that prevent reward hacking and specification gaming\n- Adversarial robustness of verifiers and red-teaming of evaluation harnesses\n- Alignment-aware verification: ensuring verifiers remain faithful as agents evolve\n\nPillar 2: Environment-Grounded Verification and Simulators\n- Faithful simulators as verification infrastructure for open-ended tasks\n- Multi-agent and self-optimizing systems for environment-grounded evaluation\n- Measuring agents in production: observability, monitoring, and runtime verification\n- Evolutionary and search-based methods for environment-driven agent optimization\n- Benchmarks and environment design that stress-test verification methods\n\nPillar 3: Diverse and Heterogeneous Verifiable Signals\n- Composing heterogeneous signals (user experience, cost/latency, calibration, multimodality) into reliable verification metrics\n- Beyond scalar rewards: holistic evaluation of agentic behavior\n- Human-in-the-loop verification and human-AI collaborative evaluation\n- Reflective and self-improving verification (agents that verify other agents)\n- Automated agent design, prompt optimization, and scaffold search with verifiable feedback\n- Formal verification of agent-generated artifacts, including the use of proof assistants and verification languages (e.g., Dafny, Rocq, and Lean)\n\nCross-Cutting Topics\n- Verification for meta-agents and Agent4Agent systems (agents that design, optimize, or evaluate other agents)\n- Self-evolving agents: stable improvement without collapse or reward hacking\n- Evaluation of agent-generated designs vs. human-engineered systems\n- Scalable oversight and verification for long-horizon, multi-step agent behavior\n- Cognitive and neuroscience-inspired verification frameworks\n\nSubmission Guidelines\nFormat: Papers should be between 4 to 9 pages (excluding references and appendices), using the NeurIPS 2026 template. We also welcome demo papers, which should be no more than 4 pages.\nDual submission policy: We welcome work that is under review or has been recently published at other venues.\nReview: Reviews will be double blind. Authors of submitted papers may be asked to contribute reviews.\nPresentation: Accepted papers will be presented as posters; select papers will be chosen for oral presentations or lightning talks.\nNon-archival: The workshop is non-archival; accepted papers will be made available on OpenReview but do not constitute formal proceedings.\nSubmissions and reviewing are handled through OpenReview.\n\nImportant Dates\nSubmission deadline: August 29, 2026\nAuthor notification: September 29, 2026\nWorkshop day: Dec 11 or 12, 2026\nThe exact workshop day (Friday, Dec 11 or Saturday, Dec 12) will be confirmed once assigned by NeurIPS. All deadlines are 23:59 Anywhere on Earth (AoE) unless otherwise noted." }, { "key": "GDDL", "title": "NeurIPS 2026 Workshop: Bridging Optimal Transport, Learning and Structured Data: Toward Geometric Distributional Learning", "subtitle": "Geometric Distributional Deep Learning (GDDL 2026) — Bridging Optimal Transport, Learning and Structured Data", "summary": "A NeurIPS 2026 workshop on Geometric Distributional Deep Learning — learning systems that jointly model the geometry and distributional nature of data, features, and representations — bringing together researchers in geometric deep learning, computational optimal transport, generative modeling, graph and manifold learning, and deep learning theory.", "cfp_full": "About the Workshop\nModern machine learning increasingly relies on representing complex data through their geometric structure or probability distribution. Geometric Deep Learning has made it possible to encode symmetries, invariances, equivariance, and relational information in non-Euclidean domains, such as graphs and manifolds, with many successful applications, ranging from protein structure prediction to neuroscience.\n\nIn parallel, viewing data or features as probability distributions has led to powerful methodologies in deep learning. In particular, optimal transport (OT) provides a natural framework for comparing, aligning, and transforming data distributions, and has contributed to notable advances in generative modeling, attention-based architectures, and representation learning.\n\nThis workshop focuses on Geometric Distributional Deep Learning: learning systems that jointly model the geometry and distributional nature of data, features and representations. Our goal is to bring together researchers in geometric deep learning, computational optimal transport, generative modeling, graph and manifold learning, and deep learning theory around a shared question: How can geometry and distributions be combined to design more scalable, efficient, and interpretable models for structured data?\n\nTopics of Interest\n- Geometry-aware Transport — Adapting optimal transport to graphs, manifolds, and physical systems\n- Neural Architectures — Processing probability distributions on graphs and manifolds with GNNs\n- Scalable Computation — Sliced methods, entropic regularization, and neural approximations at scale\n- Generative Models — Diffusion on manifolds, normalizing flows on structured spaces\n- Scientific Applications — Molecular modeling, climate prediction, neuroscience, physical sciences\n\nCall for Papers\nWe invite submissions on topics related to geometric distributional deep learning, including but not limited to:\n- Geometry-aware optimal transport methods\n- Neural architectures for distributions on non-Euclidean domains\n- Scalable computational methods for structured spaces\n- Generative models on graphs and manifolds\n- Applications in molecular modeling, climate, neuroscience\n\nSubmission Tracks\nSubmissions must follow the NeurIPS 2026 template and instructions. There will be two tracks, one for short papers (2-4 pages) and one for long papers (5-9 pages). All submissions must be anonymized. Review is double-blind via OpenReview. We welcome ongoing and unpublished works. The workshop is a non-archival venue and will not have official proceedings. Workshop submissions can be subsequently or concurrently submitted to other venues.\n- Long: 5-9 pages (excluding references)\n- Short: 2-4 pages for early-stage work\nAll accepted papers will be presented as posters. Selected papers will be invited for contributed talks.\n\nImportant Dates\n- Submission Deadline: August 29, 2026 (AoE)\n- Notification: September 29, 2026 (AoE)\n- Final Program: October 16, 2026\n- Workshop: December 12-13, 2026, Paris", "cfp_status": "published", "topics": [ "Geometry-aware Transport: adapting optimal transport to graphs, manifolds, and physical systems", "Neural Architectures: processing probability distributions on graphs and manifolds with GNNs", "Scalable Computation: sliced methods, entropic regularization, and neural approximations at scale", "Generative Models: diffusion on manifolds, normalizing flows on structured spaces", "Scientific Applications: molecular modeling, climate prediction, neuroscience, physical sciences" ], "important_dates": [ { "label": "Submission Deadline", "date": "2026-08-29" }, { "label": "Notification", "date": "2026-09-29" }, { "label": "Final Program", "date": "2026-10-16" }, { "label": "Workshop", "date": "2026-12-12" } ], "organizers": [ "Clément Bonet (Ecole Polytechnique)", "Julie Delon (ENS / Université Paris Cité)", "Nina Miolane (UC Santa Barbara)", "Youssef Mroueh (IBM Research)", "Kimia Nadjahi (CNRS / ENS)", "Justin Solomon (MIT)" ], "speakers": [ "David Alvarez-Melis (Harvard / Microsoft Research)", "Anna Calissano (University College London)", "Marco Cuturi (Apple / ENSAE-CREST)", "Stefanie Jegelka (TU Munich / MIT)", "Nicolas Keriven (CNRS / Inria)", "Soheil Kolouri (Vanderbilt University)", "Clarice Poon (University of Warwick)" ], "host_url": "https://gddl-neurips-2026.github.io", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/GDDL", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/GDDL", "location": "Paris, France", "city": "Paris", "workshop_date": "2026-12-12", "contact": "gddl.workshop.neurips2026@gmail.com", "tracks": [ { "key": "GDDL", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/GDDL", "submission_dates_raw": "" } ], "group": "science", "group_label": "AI for Science", "corpus": "NeurIPS 2026 Workshop: Bridging Optimal Transport, Learning and Structured Data: Toward Geometric Distributional Learning Geometric Distributional Deep Learning (GDDL 2026) — Bridging Optimal Transport, Learning and Structured Data A NeurIPS 2026 workshop on Geometric Distributional Deep Learning — learning systems that jointly model the geometry and distributional nature of data, features, and representations — bringing together researchers in geometric deep learning, computational optimal transport, generative modeling, graph and manifold learning, and deep learning theory. Geometry-aware Transport: adapting optimal transport to graphs, manifolds, and physical systems Neural Architectures: processing probability distributions on graphs and manifolds with GNNs Scalable Computation: sliced methods, entropic regularization, and neural approximations at scale Generative Models: diffusion on manifolds, normalizing flows on structured spaces Scientific Applications: molecular modeling, climate prediction, neuroscience, physical sciences About the Workshop\nModern machine learning increasingly relies on representing complex data through their geometric structure or probability distribution. Geometric Deep Learning has made it possible to encode symmetries, invariances, equivariance, and relational information in non-Euclidean domains, such as graphs and manifolds, with many successful applications, ranging from protein structure prediction to neuroscience.\n\nIn parallel, viewing data or features as probability distributions has led to powerful methodologies in deep learning. In particular, optimal transport (OT) provides a natural framework for comparing, aligning, and transforming data distributions, and has contributed to notable advances in generative modeling, attention-based architectures, and representation learning.\n\nThis workshop focuses on Geometric Distributional Deep Learning: learning systems that jointly model the geometry and distributional nature of data, features and representations. Our goal is to bring together researchers in geometric deep learning, computational optimal transport, generative modeling, graph and manifold learning, and deep learning theory around a shared question: How can geometry and distributions be combined to design more scalable, efficient, and interpretable models for structured data?\n\nTopics of Interest\n- Geometry-aware Transport — Adapting optimal transport to graphs, manifolds, and physical systems\n- Neural Architectures — Processing probability distributions on graphs and manifolds with GNNs\n- Scalable Computation — Sliced methods, entropic regularization, and neural approximations at scale\n- Generative Models — Diffusion on manifolds, normalizing flows on structured spaces\n- Scientific Applications — Molecular modeling, climate prediction, neuroscience, physical sciences\n\nCall for Papers\nWe invite submissions on topics related to geometric distributional deep learning, including but not limited to:\n- Geometry-aware optimal transport methods\n- Neural architectures for distributions on non-Euclidean domains\n- Scalable computational methods for structured spaces\n- Generative models on graphs and manifolds\n- Applications in molecular modeling, climate, neuroscience\n\nSubmission Tracks\nSubmissions must follow the NeurIPS 2026 template and instructions. There will be two tracks, one for short papers (2-4 pages) and one for long papers (5-9 pages). All submissions must be anonymized. Review is double-blind via OpenReview. We welcome ongoing and unpublished works. The workshop is a non-archival venue and will not have official proceedings. Workshop submissions can be subsequently or concurrently submitted to other venues.\n- Long: 5-9 pages (excluding references)\n- Short: 2-4 pages for early-stage work\nAll accepted papers will be presented as posters. Selected papers will be invited for contributed talks.\n\nImportant Dates\n- Submission Deadline: August 29, 2026 (AoE)\n- Notification: September 29, 2026 (AoE)\n- Final Program: October 16, 2026\n- Workshop: December 12-13, 2026, Paris" }, { "key": "EIML", "title": "NeurIPS 2026 Workshop: Epistemic Intelligence in Machine Learning", "subtitle": "The Roots of Intelligence", "summary": "The 3rd Workshop on Epistemic Intelligence in Machine Learning (EIML3@NeurIPS 2026) studies how learning systems recognise, represent, reason about, and act under the limits of their knowledge, aiming to establish the conceptual, representational, inferential, computational, and engineering foundations of the field.", "cfp_full": "The 3rd Workshop on Epistemic Intelligence in Machine Learning\nThe Roots of Intelligence\n12 December 2026 · Paris, France\n\nEIML@EurIPS 2025 and EIML2@ICML2026 marked the beginning of a growing community that cares about Epistemic Intelligence in Machine Learning. The conversation continues at EIML3@NeurIPS2026.\n\nWhy are we doing this?\n\nEpistemic Intelligence in Machine Learning (EIML) studies how learning systems recognise, represent, reason about, and act under the limits of their knowledge. Modern ML systems—including foundation models, generative models, autonomous agents, and AI-enabled engineering systems—are increasingly deployed in open-ended environments where failures are often epistemic: hallucination, unsafe extrapolation, brittle behaviour under distribution shift, catastrophic forgetting, overconfident decisions, and poor communication of uncertainty all arise when a system cannot distinguish what is known from what is unknown.\n\nThe first EIML workshop, held at EurIPS 2025, established the theme of epistemic intelligence around epistemic uncertainty in machine learning. The second edition, accepted at ICML 2026, broadened the agenda toward unknown unknowns, robustness, safety, alignment, foundation models, and real-world impact. The proposed NeurIPS 2026 workshop is the natural third step in the series: after establishing the community and broadening its application scope, this edition asks what must be in place for EIML to become a principled field. By foundations, we mean five coupled layers: conceptual foundations that define knowledge, ignorance, epistemic uncertainty, and epistemic agency; representational foundations that specify mathematical objects for partial knowledge; inferential foundations that determine which guarantees survive misspecification, weak prior information, partial identification, and distribution shift; computational foundations that characterise the tractability of epistemic reasoning at ML scale; and engineering foundations that govern how epistemic signals are communicated and acted upon in deployed systems.\n\nWhy now?\n\nThere is growing momentum around uncertainty-aware, reliable, and safe ML, but the relevant communities remain fragmented. Statistics studies Bayesian, frequentist, conformal, second-order, and imprecise probabilistic accounts of uncertainty. Mathematics and computer science provide formalisms for learning theory, optimisation, representation, computation, and complexity. Engineering studies how epistemic signals behave under communication, networking, robotics, and deployment constraints. Philosophy analyses knowledge, ignorance, rational belief, explanation, agency, and responsibility. These fields address overlapping problems but often use incompatible languages. Without a dedicated foundational forum, EIML risks becoming a collection of parallel literatures rather than a coherent research programme. NeurIPS is the ideal venue to crystallise shared problems, explicitly contrast frameworks, and coordinate the next phase of the field.\n\nTopics\n\n- Conceptual foundations: What is epistemic intelligence, and how does it differ from uncertainty quantification, calibration, robustness, Bayesian inference, conformal prediction, and decision theory? What does it mean for a learning system to know, not know, or act responsibly under ignorance?\n- Representations of ignorance: Which mathematical objects should represent epistemic uncertainty — sets of probability distributions, lower/upper probabilities, second-order distributions, belief functions, possibility measures, logical or modal structures, category-theoretic objects, or hybrids? What are their semantics and failure modes?\n- Statistical foundations: How should epistemic and aleatoric uncertainty be separated, measured, and validated? What guarantees remain meaningful under misspecification, partial identification, weak prior information, or distribution shift?\n- Learning theory and computation: What are the computational limits of epistemic reasoning in modern ML? Which representations are tractable at deep learning scale? Can epistemic uncertainty guide abstention, querying, exploration, safe adaptation, model revision, or fallback behaviour?\n- Engineering and communication: How should epistemic signals be embedded in AI-enabled systems, communication networks, robotics, cyber-physical systems, and human-facing interfaces? How do engineering constraints reshape the theory?\n- Evaluation and negative results: What would constitute evidence that an EIML method succeeds? Can we benchmark ignorance, blind spots, and unknown unknowns? Which existing approaches fail under carefully designed stress tests?\n\nImportant Dates\n- 29 July 2026 — Submission opens\n- 29 August 2026 — Paper submission deadline\n- 1 – 27 September 2026 — Review period\n- 29 September 2026 — Author notification\n- 18 October 2026 — Camera-ready submission deadline\n- 12 / 13 December 2026 — Main workshop", "cfp_status": "published", "topics": [ "Conceptual foundations of epistemic intelligence", "Representations of ignorance and epistemic uncertainty", "Statistical foundations (epistemic vs aleatoric uncertainty, misspecification, distribution shift)", "Learning theory and computation of epistemic reasoning", "Engineering and communication of epistemic signals", "Evaluation and negative results (benchmarking ignorance and unknown unknowns)" ], "important_dates": [ { "label": "Submission opens", "date": "2026-07-29" }, { "label": "Paper submission deadline", "date": "2026-08-29" }, { "label": "Review period", "date": "2026-09-01 to 2026-09-27" }, { "label": "Author notification", "date": "2026-09-29" }, { "label": "Camera-ready submission deadline", "date": "2026-10-18" }, { "label": "Main workshop", "date": "2026-12-12" } ], "organizers": [ "Sahar Abdelnabi (ELLIS Institute Tübingen & MPI-IS, Germany)", "Michele Caprio (University of Manchester, United Kingdom)", "Siu Lun Chau (Nanyang Technological University, Singapore)", "Julian Rodemann (CISPA Helmholtz Center for Information Security, Germany)", "Shireen Kudukkil Manchingal (Oxford Dynamics, United Kingdom)", "Krikamol Muandet (CISPA Helmholtz Center for Information Security, Germany)" ], "speakers": [ "Judith Rousseau (Université Paris Dauphine-PSL, France)", "Gitta Kutyniok (LMU Munich, Germany)", "Francesco Restuccia (Northeastern University, United States)", "Carlo Cordasco (University of Manchester, United Kingdom)" ], "host_url": "https://epistemic-intelligence-in-ml.github.io", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/EIML", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/EIML", "location": "Paris, France", "city": "Paris", "workshop_date": "2026-07-18", "contact": "pc.eiml.workshop@gmail.com", "tracks": [ { "key": "EIML", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/EIML", "submission_dates_raw": "" } ], "group": "theory", "group_label": "Theory & Methods", "corpus": "NeurIPS 2026 Workshop: Epistemic Intelligence in Machine Learning The Roots of Intelligence The 3rd Workshop on Epistemic Intelligence in Machine Learning (EIML3@NeurIPS 2026) studies how learning systems recognise, represent, reason about, and act under the limits of their knowledge, aiming to establish the conceptual, representational, inferential, computational, and engineering foundations of the field. Conceptual foundations of epistemic intelligence Representations of ignorance and epistemic uncertainty Statistical foundations (epistemic vs aleatoric uncertainty, misspecification, distribution shift) Learning theory and computation of epistemic reasoning Engineering and communication of epistemic signals Evaluation and negative results (benchmarking ignorance and unknown unknowns) The 3rd Workshop on Epistemic Intelligence in Machine Learning\nThe Roots of Intelligence\n12 December 2026 · Paris, France\n\nEIML@EurIPS 2025 and EIML2@ICML2026 marked the beginning of a growing community that cares about Epistemic Intelligence in Machine Learning. The conversation continues at EIML3@NeurIPS2026.\n\nWhy are we doing this?\n\nEpistemic Intelligence in Machine Learning (EIML) studies how learning systems recognise, represent, reason about, and act under the limits of their knowledge. Modern ML systems—including foundation models, generative models, autonomous agents, and AI-enabled engineering systems—are increasingly deployed in open-ended environments where failures are often epistemic: hallucination, unsafe extrapolation, brittle behaviour under distribution shift, catastrophic forgetting, overconfident decisions, and poor communication of uncertainty all arise when a system cannot distinguish what is known from what is unknown.\n\nThe first EIML workshop, held at EurIPS 2025, established the theme of epistemic intelligence around epistemic uncertainty in machine learning. The second edition, accepted at ICML 2026, broadened the agenda toward unknown unknowns, robustness, safety, alignment, foundation models, and real-world impact. The proposed NeurIPS 2026 workshop is the natural third step in the series: after establishing the community and broadening its application scope, this edition asks what must be in place for EIML to become a principled field. By foundations, we mean five coupled layers: conceptual foundations that define knowledge, ignorance, epistemic uncertainty, and epistemic agency; representational foundations that specify mathematical objects for partial knowledge; inferential foundations that determine which guarantees survive misspecification, weak prior information, partial identification, and distribution shift; computational foundations that characterise the tractability of epistemic reasoning at ML scale; and engineering foundations that govern how epistemic signals are communicated and acted upon in deployed systems.\n\nWhy now?\n\nThere is growing momentum around uncertainty-aware, reliable, and safe ML, but the relevant communities remain fragmented. Statistics studies Bayesian, frequentist, conformal, second-order, and imprecise probabilistic accounts of uncertainty. Mathematics and computer science provide formalisms for learning theory, optimisation, representation, computation, and complexity. Engineering studies how epistemic signals behave under communication, networking, robotics, and deployment constraints. Philosophy analyses knowledge, ignorance, rational belief, explanation, agency, and responsibility. These fields address overlapping problems but often use incompatible languages. Without a dedicated foundational forum, EIML risks becoming a collection of parallel literatures rather than a coherent research programme. NeurIPS is the ideal venue to crystallise shared problems, explicitly contrast frameworks, and coordinate the next phase of the field.\n\nTopics\n\n- Conceptual foundations: What is epistemic intelligence, and how does it differ from uncertainty quantification, calibration, robustness, Bayesian inference, conformal prediction, and decision theory? What does it mean for a learning system to know, not know, or act responsibly under ignorance?\n- Representations of ignorance: Which mathematical objects should represent epistemic uncertainty — sets of probability distributions, lower/upper probabilities, second-order distributions, belief functions, possibility measures, logical or modal structures, category-theoretic objects, or hybrids? What are their semantics and failure modes?\n- Statistical foundations: How should epistemic and aleatoric uncertainty be separated, measured, and validated? What guarantees remain meaningful under misspecification, partial identification, weak prior information, or distribution shift?\n- Learning theory and computation: What are the computational limits of epistemic reasoning in modern ML? Which representations are tractable at deep learning scale? Can epistemic uncertainty guide abstention, querying, exploration, safe adaptation, model revision, or fallback behaviour?\n- Engineering and communication: How should epistemic signals be embedded in AI-enabled systems, communication networks, robotics, cyber-physical systems, and human-facing interfaces? How do engineering constraints reshape the theory?\n- Evaluation and negative results: What would constitute evidence that an EIML method succeeds? Can we benchmark ignorance, blind spots, and unknown unknowns? Which existing approaches fail under carefully designed stress tests?\n\nImportant Dates\n- 29 July 2026 — Submission opens\n- 29 August 2026 — Paper submission deadline\n- 1 – 27 September 2026 — Review period\n- 29 September 2026 — Author notification\n- 18 October 2026 — Camera-ready submission deadline\n- 12 / 13 December 2026 — Main workshop" }, { "key": "FMTS", "title": "NeurIPS 2026 Workshop: Foundation Models for Temporal Systems From Forecasting to World Modeling", "subtitle": "FMTS 2026", "summary": "A workshop bringing forecasting, simulation, multimodal temporal data, and reliability into one research agenda for foundation models of evolving real-world systems, organized around four connected axes: forecasting and simulation tasks, temporal data and environments, temporal models, and evaluation and reliability.", "cfp_full": "Why this workshop\n\nFrom forecasts to models of evolving systems\n\nTemporal machine learning is moving beyond task-specific prediction. The next challenge is to build foundation models that can forecast, simulate, adapt, and support decisions in systems that evolve across time, modalities, and scales.\n\nTemporal world modeling\n\nReal systems are multimodal, asynchronous, event-driven, multi-scale, and subject to distribution shift. Treating a model as more than a forecaster changes what matters: calibrated uncertainty, long-horizon consistency, simulation fidelity, causal structure, and reliable operation with retrieval, tools, and external context.\n\nWorkshop framing\n\nA shared model for forecasts, rollouts, and decisions. The workshop connects multimodal histories and interventions to models that maintain temporal state, imagine alternative futures, and remain trustworthy as conditions change. A common systems view connects the workshop's four research axes: tasks, data and environments, models, and evaluation.\n\nResearch agenda - Four connected research axes\n\nThe workshop connects task design, data and environments, model architecture, and evaluation. Datasets, simulators, and temporal environments are treated as primary research artifacts, not supporting material.\n\nAXIS 01 - Forecasting and simulation tasks: Long-horizon and multi-resolution prediction, multimodal contextual forecasting, calibrated probabilistic forecasts, trajectory simulation, and adaptive or agentic forecasting systems. Includes forecasting from sparse observations and retrieval-augmented or agentic approaches.\n\nAXIS 02 - Temporal data and environments: Rich environments combining time series with text, video, graphs, sensors, trajectories, actions, and events, alongside realistic benchmarks and scalable simulation suites. Includes large-scale pretraining corpora, physical evaluation suites, synthetic data, and benchmark realism.\n\nAXIS 03 - Temporal models: Irregular and event-based models, hierarchical multi-timescale systems, state-space architectures, generative temporal models, and time-series foundation models. Includes time-aware architectures for irregular sampling and multimodal fusion with embodied systems.\n\nAXIS 04 - Evaluation and reliability: Leakage-aware evaluation, robustness under shift, calibration, long-horizon and simulation consistency, neural scaling laws, contamination audits, and reproducibility.\n\nEncouraged Contributions\n- Datasets, benchmarks, and simulators\n- Negative results and replication studies\n- Production deployment and operational reports\n\nThe workshop accepts complete results, work in progress, position papers, and preliminary or negative findings alike.\n\nSubmission Format\n\nSubmissions are up to 4 pages, excluding references and appendices, with double-blind review and no separate archival proceedings.\n\nReview Process\n\nEach paper receives three reviews via OpenReview. Selection emphasizes technical quality, topical diversity, and interaction across workshop communities, with priority given to junior researchers for spotlight slots.\n\nImportant dates\n\n- Submission deadline: August 29, 2026, 11:59 pm AoE\n- Author notification: September 29, 2026 (fixed by NeurIPS)\n- Camera-ready: November 6, 2026 (tentative)\n- Workshop: December 11-12, 2026 (exact day and room TBA)\n\nAll submission deadlines are at 11:59 pm AoE. The notification date is set by the NeurIPS workshop chairs and cannot be extended.", "cfp_status": "published", "topics": [ "Forecasting and simulation tasks: long-horizon and multi-resolution prediction, multimodal contextual forecasting, calibrated probabilistic forecasts, trajectory simulation, adaptive/agentic forecasting", "Temporal data and environments: environments combining time series with text, video, graphs, sensors, trajectories, actions, and events; realistic benchmarks and scalable simulation suites", "Temporal models: irregular and event-based models, hierarchical multi-timescale systems, state-space architectures, generative temporal models, time-series foundation models", "Evaluation and reliability: leakage-aware evaluation, robustness under shift, calibration, long-horizon and simulation consistency, neural scaling laws, contamination audits, reproducibility" ], "important_dates": [ { "label": "Submission deadline", "date": "2026-08-29" }, { "label": "Author notification", "date": "2026-09-29" }, { "label": "Camera-ready", "date": "2026-11-06" }, { "label": "Workshop", "date": "2026-12-11" } ], "organizers": [ "Boris N. Oreshkin (Amazon)", "Mayank Jauhari (Amazon)", "Danielle Maddix Robinson (Siemens Physical AI)", "Omri Azencot (Ben-Gurion University)", "Ming Jin (Griffith University)", "Emadeldeen Eldele (Khalifa University)", "Chenghao Liu (Datadog)", "N. Benjamin Erichson (Berkeley Lab / ICSI)" ], "speakers": [ "Rose Yu (UC San Diego)", "Michael W. Mahoney (UC Berkeley / ICSI / LBNL)", "Abdul Fatir Ansari (AWS)", "Aditi Krishnapriyan (UC Berkeley)", "Marinka Zitnik (Harvard Medical School)", "Daniel F. Schmidt (Monash University)", "Flora Salim (UNSW Sydney)" ], "host_url": "https://fmts-workshop.github.io/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/FMTS", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/FMTS", "location": "Sydney, Australia", "city": "Sydney", "workshop_date": "2026-12-11", "contact": "fmtsworkshop@gmail.com", "tracks": [ { "key": "FMTS", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/FMTS", "submission_dates_raw": "" } ], "group": "fm", "group_label": "Foundation Models & LLMs", "corpus": "NeurIPS 2026 Workshop: Foundation Models for Temporal Systems From Forecasting to World Modeling FMTS 2026 A workshop bringing forecasting, simulation, multimodal temporal data, and reliability into one research agenda for foundation models of evolving real-world systems, organized around four connected axes: forecasting and simulation tasks, temporal data and environments, temporal models, and evaluation and reliability. Forecasting and simulation tasks: long-horizon and multi-resolution prediction, multimodal contextual forecasting, calibrated probabilistic forecasts, trajectory simulation, adaptive/agentic forecasting Temporal data and environments: environments combining time series with text, video, graphs, sensors, trajectories, actions, and events; realistic benchmarks and scalable simulation suites Temporal models: irregular and event-based models, hierarchical multi-timescale systems, state-space architectures, generative temporal models, time-series foundation models Evaluation and reliability: leakage-aware evaluation, robustness under shift, calibration, long-horizon and simulation consistency, neural scaling laws, contamination audits, reproducibility Why this workshop\n\nFrom forecasts to models of evolving systems\n\nTemporal machine learning is moving beyond task-specific prediction. The next challenge is to build foundation models that can forecast, simulate, adapt, and support decisions in systems that evolve across time, modalities, and scales.\n\nTemporal world modeling\n\nReal systems are multimodal, asynchronous, event-driven, multi-scale, and subject to distribution shift. Treating a model as more than a forecaster changes what matters: calibrated uncertainty, long-horizon consistency, simulation fidelity, causal structure, and reliable operation with retrieval, tools, and external context.\n\nWorkshop framing\n\nA shared model for forecasts, rollouts, and decisions. The workshop connects multimodal histories and interventions to models that maintain temporal state, imagine alternative futures, and remain trustworthy as conditions change. A common systems view connects the workshop's four research axes: tasks, data and environments, models, and evaluation.\n\nResearch agenda - Four connected research axes\n\nThe workshop connects task design, data and environments, model architecture, and evaluation. Datasets, simulators, and temporal environments are treated as primary research artifacts, not supporting material.\n\nAXIS 01 - Forecasting and simulation tasks: Long-horizon and multi-resolution prediction, multimodal contextual forecasting, calibrated probabilistic forecasts, trajectory simulation, and adaptive or agentic forecasting systems. Includes forecasting from sparse observations and retrieval-augmented or agentic approaches.\n\nAXIS 02 - Temporal data and environments: Rich environments combining time series with text, video, graphs, sensors, trajectories, actions, and events, alongside realistic benchmarks and scalable simulation suites. Includes large-scale pretraining corpora, physical evaluation suites, synthetic data, and benchmark realism.\n\nAXIS 03 - Temporal models: Irregular and event-based models, hierarchical multi-timescale systems, state-space architectures, generative temporal models, and time-series foundation models. Includes time-aware architectures for irregular sampling and multimodal fusion with embodied systems.\n\nAXIS 04 - Evaluation and reliability: Leakage-aware evaluation, robustness under shift, calibration, long-horizon and simulation consistency, neural scaling laws, contamination audits, and reproducibility.\n\nEncouraged Contributions\n- Datasets, benchmarks, and simulators\n- Negative results and replication studies\n- Production deployment and operational reports\n\nThe workshop accepts complete results, work in progress, position papers, and preliminary or negative findings alike.\n\nSubmission Format\n\nSubmissions are up to 4 pages, excluding references and appendices, with double-blind review and no separate archival proceedings.\n\nReview Process\n\nEach paper receives three reviews via OpenReview. Selection emphasizes technical quality, topical diversity, and interaction across workshop communities, with priority given to junior researchers for spotlight slots.\n\nImportant dates\n\n- Submission deadline: August 29, 2026, 11:59 pm AoE\n- Author notification: September 29, 2026 (fixed by NeurIPS)\n- Camera-ready: November 6, 2026 (tentative)\n- Workshop: December 11-12, 2026 (exact day and room TBA)\n\nAll submission deadlines are at 11:59 pm AoE. The notification date is set by the NeurIPS workshop chairs and cannot be extended." }, { "key": "PALM", "title": "NeurIPS 2026 Workshop: Personalized, Aligned, Long-Term Memory for AI Systems", "subtitle": "NeurIPS 2026 Workshop PALM", "summary": "PALM is a workshop on safe, personalized, and long-term memory for AI agents and conversational assistants, bringing together work on memory architectures, personalization, multimodal and embodied memory, agentic memory, cognitive-inspired memory, evaluation, and safety.", "cfp_full": "About the Workshop\n\nLong-term memory is becoming a core component of modern AI systems. Conversational assistants and agentic systems are increasingly expected to retain information across sessions, personalize to users, reason over long horizons, and act consistently across tasks, tools, and modalities. This shift moves AI systems beyond single-context interactions toward persistent memory layers that encode, retrieve, update, and sometimes forget past experience.\n\nPALM is a workshop on safe, personalized, and long-term memory for AI agents and conversational assistants. The workshop will bring together researchers working on memory architectures, personalization, multimodal and embodied memory, agentic and multi-agent memory, cognitive and neuroscience-inspired memory, evaluation, and safety. Our goal is to build a shared research agenda around how AI systems should remember, what they should forget, how memory should be evaluated, and how persistent memory can be made controllable, transparent, and safe.\n\nTopics of Interest\n\nTopics of interest include, but are not limited to:\n\n01 Memory Architectures for Conversational Assistants\nWe welcome work on how conversational assistants should write, retrieve, update, consolidate, and forget memories across sessions. Example topics include memory stores for multi-session dialogue, retrieval policies for user preferences, memory consolidation from conversation histories, and mechanisms for handling stale or contradictory memories.\n\n02 Memory for LLM Agents & Multi-Agent Systems\nWe invite work on how agents use memory to support long-horizon tasks, tool use, collaboration, and coordination. Example topics include persistent task histories, tool-use traces, shared memory across agents, memory provenance, memory isolation between agents, and mechanisms for propagating or restricting memories in multi-agent systems.\n\n03 Multimodal, Visual, Video & Embodied Memory\nWe encourage submissions on memory systems that operate across text, images, video, audio, sensorimotor streams, robotics, and embodied environments. Example topics include long-term video memory, visual retrieval for agents, spatial memory for embodied systems, multimodal event memory, and memory for AR/VR or robotic assistants.\n\n04 Neuroscience-Inspired & Cognitive Memory Models\nWe welcome work that draws inspiration from human and biological memory systems to inform AI memory design. Example topics include complementary learning systems, episodic-to-semantic consolidation, replay, forgetting, abstraction, cognitive maps, and comparisons between human and machine memory limitations or biases.\n\n05 Benchmarking & Evaluation\nWe invite work on how to evaluate long-term memory systems beyond short-context recall. Example topics include long-horizon memory benchmarks, temporal reasoning over past events, memory update and deletion tests, contradiction handling, abstention under uncertainty, oracle-retrieval comparisons, and evaluations of real-world memory competence.\n\n06 Safety, Privacy & Security\nWe encourage work on the risks introduced by persistent memory and memory-enabled personalization. Example topics include memory poisoning, prompt injection through stored memories, sleeper memories, privacy leakage, cross-domain leakage, sycophancy, harmful belief reinforcement, long-term manipulation, and alignment drift.\n\n07 User Control & Transparency\nWe welcome work on how users can understand and control what AI systems remember. Example topics include interfaces for inspecting, editing, deleting, and scoping memories; consent and access-control mechanisms; memory provenance; right-to-be-forgotten mechanisms; and human-centered evaluations of memory transparency.\n\nCall for Papers\n\nWe invite submissions on long-term memory for AI agents, conversational assistants, and personalized AI systems. We welcome work from machine learning, NLP, AI agents, HCI, cognitive science, neuroscience, privacy, security, and AI safety. Submissions may present new architectures, benchmarks, datasets, evaluations, systems, theoretical perspectives, position papers, negative results, or interdisciplinary analyses.\n\nSubmission tracks\nFull-Length Papers: Up to 9 pages (excluding references and supplementary materials).\nShort Papers: Up to 4 pages (excluding references and supplementary materials).\n\nFormat & policies\nStyle files and templates: To prepare your submission to PALM Workshop 2026, please use the NeurIPS 2026 template.\nDual-submission policy: The workshop will adopt a non-archival policy, welcoming ongoing and unpublished work, as well as papers under review or recently accepted at other venues (provided they do not breach dual-submission or anonymity policies of the other venue). Workshop submissions can be subsequently or concurrently submitted to other venues.\nVisibility: Accepted papers will be made public, but rejected submissions and reviews will not.\nDouble-blind reviewing: Submissions must be fully anonymized. This policy applies to any supplementary or linked material as well, including code. Any papers found to be in violation of this policy may be desk-rejected.\nLLM usage policy: AI-generated papers are not allowed. AI assistance is permitted, but submissions must be primarily human-authored, reflecting original thought and analysis.\n\nImportant Dates\nAugust 24, 2026 - Workshop paper submission deadline\nSeptember 29, 2026 - Notification of acceptance\nTBD - Camera-ready\nDecember 12 or 13, 2026 - Workshop at NeurIPS 2026, Paris (exact day TBA)\nAll deadlines are 11:59pm AoE (Anywhere on Earth).", "cfp_status": "published", "topics": [ "Memory Architectures for Conversational Assistants", "Memory for LLM Agents & Multi-Agent Systems", "Multimodal, Visual, Video & Embodied Memory", "Neuroscience-Inspired & Cognitive Memory Models", "Benchmarking & Evaluation of long-term memory systems", "Safety, Privacy & Security of persistent memory", "User Control & Transparency over AI memory" ], "important_dates": [ { "label": "Workshop paper submission", "date": "2026-08-24" }, { "label": "Notification of acceptance", "date": "2026-09-29" }, { "label": "Camera-ready", "date": "TBD" }, { "label": "Workshop", "date": "December 12 or 13, 2026" } ], "organizers": [ "Mario Fritz (CISPA Helmholtz Center)", "Seong Joon Oh (KAIST)", "Sahar Abdelnabi (ELLIS Institute Tübingen & MPI-IS)", "Shawn Shen (Memories.ai & Univ. of Bristol)", "Hugo D. Lopes (Google DeepMind)", "Haritz Puerto (ELLIS Institute Tübingen & MPI-IS)", "Ivaxi Sheth (CISPA Helmholtz Center)", "Seokwon Jung (KAIST)" ], "speakers": [ "Weiwen Liu (Shanghai Jiao Tong University)" ], "host_url": "https://palm-neurips-2026.github.io/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/PALM", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/PALM", "location": "Paris, France", "city": "Paris", "workshop_date": "2026-12-11", "contact": "palm-neurips-2026@googlegroups.com", "tracks": [ { "key": "PALM", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/PALM", "submission_dates_raw": "" } ], "group": "fm", "group_label": "Foundation Models & LLMs", "corpus": "NeurIPS 2026 Workshop: Personalized, Aligned, Long-Term Memory for AI Systems NeurIPS 2026 Workshop PALM PALM is a workshop on safe, personalized, and long-term memory for AI agents and conversational assistants, bringing together work on memory architectures, personalization, multimodal and embodied memory, agentic memory, cognitive-inspired memory, evaluation, and safety. Memory Architectures for Conversational Assistants Memory for LLM Agents & Multi-Agent Systems Multimodal, Visual, Video & Embodied Memory Neuroscience-Inspired & Cognitive Memory Models Benchmarking & Evaluation of long-term memory systems Safety, Privacy & Security of persistent memory User Control & Transparency over AI memory About the Workshop\n\nLong-term memory is becoming a core component of modern AI systems. Conversational assistants and agentic systems are increasingly expected to retain information across sessions, personalize to users, reason over long horizons, and act consistently across tasks, tools, and modalities. This shift moves AI systems beyond single-context interactions toward persistent memory layers that encode, retrieve, update, and sometimes forget past experience.\n\nPALM is a workshop on safe, personalized, and long-term memory for AI agents and conversational assistants. The workshop will bring together researchers working on memory architectures, personalization, multimodal and embodied memory, agentic and multi-agent memory, cognitive and neuroscience-inspired memory, evaluation, and safety. Our goal is to build a shared research agenda around how AI systems should remember, what they should forget, how memory should be evaluated, and how persistent memory can be made controllable, transparent, and safe.\n\nTopics of Interest\n\nTopics of interest include, but are not limited to:\n\n01 Memory Architectures for Conversational Assistants\nWe welcome work on how conversational assistants should write, retrieve, update, consolidate, and forget memories across sessions. Example topics include memory stores for multi-session dialogue, retrieval policies for user preferences, memory consolidation from conversation histories, and mechanisms for handling stale or contradictory memories.\n\n02 Memory for LLM Agents & Multi-Agent Systems\nWe invite work on how agents use memory to support long-horizon tasks, tool use, collaboration, and coordination. Example topics include persistent task histories, tool-use traces, shared memory across agents, memory provenance, memory isolation between agents, and mechanisms for propagating or restricting memories in multi-agent systems.\n\n03 Multimodal, Visual, Video & Embodied Memory\nWe encourage submissions on memory systems that operate across text, images, video, audio, sensorimotor streams, robotics, and embodied environments. Example topics include long-term video memory, visual retrieval for agents, spatial memory for embodied systems, multimodal event memory, and memory for AR/VR or robotic assistants.\n\n04 Neuroscience-Inspired & Cognitive Memory Models\nWe welcome work that draws inspiration from human and biological memory systems to inform AI memory design. Example topics include complementary learning systems, episodic-to-semantic consolidation, replay, forgetting, abstraction, cognitive maps, and comparisons between human and machine memory limitations or biases.\n\n05 Benchmarking & Evaluation\nWe invite work on how to evaluate long-term memory systems beyond short-context recall. Example topics include long-horizon memory benchmarks, temporal reasoning over past events, memory update and deletion tests, contradiction handling, abstention under uncertainty, oracle-retrieval comparisons, and evaluations of real-world memory competence.\n\n06 Safety, Privacy & Security\nWe encourage work on the risks introduced by persistent memory and memory-enabled personalization. Example topics include memory poisoning, prompt injection through stored memories, sleeper memories, privacy leakage, cross-domain leakage, sycophancy, harmful belief reinforcement, long-term manipulation, and alignment drift.\n\n07 User Control & Transparency\nWe welcome work on how users can understand and control what AI systems remember. Example topics include interfaces for inspecting, editing, deleting, and scoping memories; consent and access-control mechanisms; memory provenance; right-to-be-forgotten mechanisms; and human-centered evaluations of memory transparency.\n\nCall for Papers\n\nWe invite submissions on long-term memory for AI agents, conversational assistants, and personalized AI systems. We welcome work from machine learning, NLP, AI agents, HCI, cognitive science, neuroscience, privacy, security, and AI safety. Submissions may present new architectures, benchmarks, datasets, evaluations, systems, theoretical perspectives, position papers, negative results, or interdisciplinary analyses.\n\nSubmission tracks\nFull-Length Papers: Up to 9 pages (excluding references and supplementary materials).\nShort Papers: Up to 4 pages (excluding references and supplementary materials).\n\nFormat & policies\nStyle files and templates: To prepare your submission to PALM Workshop 2026, please use the NeurIPS 2026 template.\nDual-submission policy: The workshop will adopt a non-archival policy, welcoming ongoing and unpublished work, as well as papers under review or recently accepted at other venues (provided they do not breach dual-submission or anonymity policies of the other venue). Workshop submissions can be subsequently or concurrently submitted to other venues.\nVisibility: Accepted papers will be made public, but rejected submissions and reviews will not.\nDouble-blind reviewing: Submissions must be fully anonymized. This policy applies to any supplementary or linked material as well, including code. Any papers found to be in violation of this policy may be desk-rejected.\nLLM usage policy: AI-generated papers are not allowed. AI assistance is permitted, but submissions must be primarily human-authored, reflecting original thought and analysis.\n\nImportant Dates\nAugust 24, 2026 - Workshop paper submission deadline\nSeptember 29, 2026 - Notification of acceptance\nTBD - Camera-ready\nDecember 12 or 13, 2026 - Workshop at NeurIPS 2026, Paris (exact day TBA)\nAll deadlines are 11:59pm AoE (Anywhere on Earth)." }, { "key": "Pre-to-Post", "title": "NeurIPS 2026 Workshop: Transitioning from Pre-Training to Post-Training", "subtitle": "Pre-to-Post NeurIPS 2026", "summary": "A NeurIPS 2026 workshop on the interaction between pre-training and post-training for large language models — what a base model must possess for post-training to succeed, the roles and properties of each training stage, and how to predict success or failure.", "cfp_full": "Transitioning from Pre-Training to Post-Training\nA workshop on the interaction between pre-training and post-training for LLM training. Key themes: (i) what a base model must possess for post-training to succeed, (ii) what are the roles and properties for each stage of training, and (iii) how do we predict success or failure.\n\nOverview\nModern foundation models are built in stages: large-scale pre-training, followed by increasingly complex post-training: supervised fine-tuning, preference optimization, reinforcement learning, self-improvement, and (on-policy) distillation. Yet we lack a principled understanding of how these stages relate: what each stage is for, what a base model must provide for the next to succeed, and which steps are actually necessary. As instruction and reasoning data increasingly enters the pre-training mix, the boundary between “pre” and “post” is itself becoming blurry.\n\nThis workshop brings together theory, empirical evidence, and benchmarks to turn folklore about this pipeline into science.\n\nCentral questions\n- What must a pre-trained model possess for post-training to succeed — and how do we even define “success”?\n- How do different post-training procedures reshape the base model, beyond targeted benchmark gains?\n- When and why does post-training fail, and can pre-training-side signals predict it?\n- How do data and optimization choices govern the transitions between stages?\n\nTopics of Interest\n- Foundations laid during pre-training: how data mixtures, curricula, continued or mid-training, learning-rate decay, and other late-stage pre-training decisions shape downstream capabilities.\n- The mechanics of post-training: comparing supervised fine-tuning, reinforcement learning from human or AI feedback, reinforcement learning with verifiable rewards, and distillation.\n- Development of model behaviors: identifying when alignment, reasoning, instruction following, refusal, persona, and other behaviors emerge during specific stages of post-training.\n- Interactions between pre-training and post-training data: how particular pre-training data mixtures, domains, curricula, or objectives make subsequent post-training more or less effective.\n- Failure modes and fundamental limits: mode or entropy collapse, reward hacking, capability forgetting, alignment taxes.\n- Data and optimization across the training transition: synthetic data, scaling laws for supervised, preference, and reinforcement-learning data.\n- Predicting post-training outcomes: metrics, representations, or behavioral signals during pre-training that forecast later trainability.\n- Reimagining the training pipeline: folding traditionally post-training data and objectives into pre-training, jointly designing training stages.\n- Evaluation and open science: causal experiments, standardized protocols, intermediate checkpoints, and openly reproducible training studies.\n\nSubmission Guidelines\nFull-day, in-person workshop. Non-archival short & long papers, formatted in NeurIPS paper style, hosted on OpenReview. Short papers: 4–5 pages. Long papers: the chosen format's main-conference page limit. Page limits exclude references and appendices for both tracks. Each submission must nominate a reciprocal reviewer, who may be contacted to review if additional reviewers are needed. Submissions are non-archival; work already published at NeurIPS or other major ML conferences is not eligible.\n\nImportant dates\n- Submission portal opens: August 1, 2026\n- Submission deadline: August 29, 2026, 11:59 PM AoE\n- Author notification: September 29, 2026, 11:59 PM AoE\n- Workshop: December 11, 2026 (Sydney, Australia)", "cfp_status": "published", "topics": [ "Foundations laid during pre-training (data mixtures, curricula, mid-training, learning-rate decay)", "The mechanics of post-training (SFT, RLHF/RLAIF, RLVR, distillation)", "Development of model behaviors (alignment, reasoning, instruction following, refusal, persona)", "Interactions between pre-training and post-training data", "Failure modes and fundamental limits (mode/entropy collapse, reward hacking, forgetting, alignment taxes)", "Data and optimization across the training transition (synthetic data, scaling laws)", "Predicting post-training outcomes from pre-training signals", "Reimagining the training pipeline", "Evaluation and open science" ], "important_dates": [ { "label": "Submission Portal Opens", "date": "2026-08-01" }, { "label": "Submission Deadline", "date": "2026-08-29" }, { "label": "Author Notification", "date": "2026-09-29" }, { "label": "Workshop", "date": "2026-12-11" } ], "organizers": [], "speakers": [], "host_url": "https://pretrain2posttrain.github.io", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/Pre-to-Post", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/Pre-to-Post", "location": "Sydney, Australia", "city": "Sydney", "workshop_date": "2026-12-10", "contact": "pre2post-neurips2026@googlegroups.com", "tracks": [ { "key": "Pre-to-Post", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/Pre-to-Post", "submission_dates_raw": "" } ], "group": "fm", "group_label": "Foundation Models & LLMs", "corpus": "NeurIPS 2026 Workshop: Transitioning from Pre-Training to Post-Training Pre-to-Post NeurIPS 2026 A NeurIPS 2026 workshop on the interaction between pre-training and post-training for large language models — what a base model must possess for post-training to succeed, the roles and properties of each training stage, and how to predict success or failure. Foundations laid during pre-training (data mixtures, curricula, mid-training, learning-rate decay) The mechanics of post-training (SFT, RLHF/RLAIF, RLVR, distillation) Development of model behaviors (alignment, reasoning, instruction following, refusal, persona) Interactions between pre-training and post-training data Failure modes and fundamental limits (mode/entropy collapse, reward hacking, forgetting, alignment taxes) Data and optimization across the training transition (synthetic data, scaling laws) Predicting post-training outcomes from pre-training signals Reimagining the training pipeline Evaluation and open science Transitioning from Pre-Training to Post-Training\nA workshop on the interaction between pre-training and post-training for LLM training. Key themes: (i) what a base model must possess for post-training to succeed, (ii) what are the roles and properties for each stage of training, and (iii) how do we predict success or failure.\n\nOverview\nModern foundation models are built in stages: large-scale pre-training, followed by increasingly complex post-training: supervised fine-tuning, preference optimization, reinforcement learning, self-improvement, and (on-policy) distillation. Yet we lack a principled understanding of how these stages relate: what each stage is for, what a base model must provide for the next to succeed, and which steps are actually necessary. As instruction and reasoning data increasingly enters the pre-training mix, the boundary between “pre” and “post” is itself becoming blurry.\n\nThis workshop brings together theory, empirical evidence, and benchmarks to turn folklore about this pipeline into science.\n\nCentral questions\n- What must a pre-trained model possess for post-training to succeed — and how do we even define “success”?\n- How do different post-training procedures reshape the base model, beyond targeted benchmark gains?\n- When and why does post-training fail, and can pre-training-side signals predict it?\n- How do data and optimization choices govern the transitions between stages?\n\nTopics of Interest\n- Foundations laid during pre-training: how data mixtures, curricula, continued or mid-training, learning-rate decay, and other late-stage pre-training decisions shape downstream capabilities.\n- The mechanics of post-training: comparing supervised fine-tuning, reinforcement learning from human or AI feedback, reinforcement learning with verifiable rewards, and distillation.\n- Development of model behaviors: identifying when alignment, reasoning, instruction following, refusal, persona, and other behaviors emerge during specific stages of post-training.\n- Interactions between pre-training and post-training data: how particular pre-training data mixtures, domains, curricula, or objectives make subsequent post-training more or less effective.\n- Failure modes and fundamental limits: mode or entropy collapse, reward hacking, capability forgetting, alignment taxes.\n- Data and optimization across the training transition: synthetic data, scaling laws for supervised, preference, and reinforcement-learning data.\n- Predicting post-training outcomes: metrics, representations, or behavioral signals during pre-training that forecast later trainability.\n- Reimagining the training pipeline: folding traditionally post-training data and objectives into pre-training, jointly designing training stages.\n- Evaluation and open science: causal experiments, standardized protocols, intermediate checkpoints, and openly reproducible training studies.\n\nSubmission Guidelines\nFull-day, in-person workshop. Non-archival short & long papers, formatted in NeurIPS paper style, hosted on OpenReview. Short papers: 4–5 pages. Long papers: the chosen format's main-conference page limit. Page limits exclude references and appendices for both tracks. Each submission must nominate a reciprocal reviewer, who may be contacted to review if additional reviewers are needed. Submissions are non-archival; work already published at NeurIPS or other major ML conferences is not eligible.\n\nImportant dates\n- Submission portal opens: August 1, 2026\n- Submission deadline: August 29, 2026, 11:59 PM AoE\n- Author notification: September 29, 2026, 11:59 PM AoE\n- Workshop: December 11, 2026 (Sydney, Australia)" }, { "key": "CODEC-FM", "title": "NeurIPS Workshop on Collaborative, Open, and Decentralized Training of Foundation Models", "subtitle": "CODEC-FM 2026", "summary": "CODEC-FM brings together the open-model and decentralized-training communities to address scaling foundation model training across distributed, heterogeneous compute — from inter-datacenter training to internet-scale volunteer collaboration.", "cfp_full": "About\n\nTraining large-scale foundation models today depends on massive, centralized GPU clusters that are inaccessible to most academic institutions, startups, and industries.\n\nThis concentration of compute creates high barriers to entry, centralizes AI innovation, and limits broader progress in developing and studying frontier-scale foundation models. Decentralization and resource pooling provide a way to enable large-scale runs without these barriers, by training models across geographically distributed and heterogeneous devices — from coordinated inter-datacenter settings to consumer devices collaborating over the internet.\n\nThis workshop brings together researchers and practitioners from the open-model and decentralized-training communities to address the core technical challenges of this paradigm, extending foundation model training beyond the confines of a single datacenter and supporting more scalable, collaborative open model development.\n\nTopics of interest\n- Open model development and community infrastructure\n- Communication-efficient distributed training\n- Asynchronous and local-update optimization\n- Inter-datacenter training\n- Internet-scale volunteer collaboration\n- Federated learning at scale\n- Heterogeneous devices and interconnects\n- Fault tolerance, stragglers, and elasticity\n- Parallelism strategies over low-bandwidth interconnects\n- Security and robustness in trustless settings\n- Incentives, governance, and economic models\n- Benchmarks and evaluation\n\nImportant Dates (All deadlines are 23:59 anywhere on Earth):\n- Submission deadline: August 29, 2026\n- Reviewing deadline: September 19, 2026\n- Notification of acceptance: September 29, 2026\n- Camera-ready deadline: TBA\n- Workshop day: December 11 or 12, 2026\n\nCall for Papers\n\nSubmission format\n- Long papers: up to 6 pages — eligible for oral presentation\n- Tiny papers: up to 2 pages — poster presentations only\n- Page limits exclude references and appendix\n- Use the official NeurIPS 2026 LaTeX style files (linked from the NeurIPS 2026 Call for Papers)\n- All submissions are non-archival\n- All authors are required to have a valid OpenReview profile\n- Submissions are managed through OpenReview\n\nWhat we accept\n- New research results\n- Position papers\n- Systems and infrastructure descriptions\n- Benchmark papers\n- Negative findings that clarify trade-offs\n\nTopics we accept\n\nWe encourage contributions addressing the core challenges of collaborative, open, and decentralized training, including but not limited to:\n- Open model development and community infrastructure: systems, architectures, algorithms, and governance mechanisms that enable large-scale collaborative open model development on geographically distributed compute.\n- Communication-efficient and asynchronous training: compression, scheduling, and optimization techniques that enable large-scale training across all parallelism axes over low-bandwidth and high-latency interconnects, while improving device utilization through asynchronous execution.\n- Heterogeneous and fault-tolerant systems: training algorithms and systems that support heterogeneous devices and interconnects in decentralized networks, maximize aggregate throughput, and remain robust to unreliable nodes, stragglers, and intermittent connectivity.\n- Security and robustness: methods for detecting malicious or faulty participants, ensuring robust convergence, and enabling secure collaboration in trustless or partially trusted decentralized training environments.\n\nReview process\n\nReviewing is double blind. Each submission will receive three reviews. Our program committee includes international experts with reviewing experience at top-tier conferences and recent workshops such as MCDC@ICLR'25 and FL@FM-NeurIPS'24. Organizers follow the NeurIPS conflict-of-interest policy. Six accepted papers will be selected for oral presentation, and all accepted papers will be presented in the poster session.\n\nAwards\n\nWe will present awards for the best paper, best student paper, and best tiny paper. Specific details will be communicated closer to the workshop date.\n\nCall for reviewers\n\nWe are recruiting reviewers from the broader community. If you work on distributed or decentralized training, federated learning, communication-efficient optimization, ML systems, or security for collaborative learning, we would be grateful for your help. To volunteer, fill out the reviewer sign-up form.\n\nDual submission\n\nBecause this workshop is non-archival, concurrent submission to other venues is permitted (subject to those venues' policies), and accepting a paper here does not preclude its later publication elsewhere. Accepted papers and the workshop schedule will be made publicly available on this website, and talks will be made public after the workshop.\n\nCode of conduct\n\nAll authors, reviewers, and attendees are expected to adhere to the NeurIPS Code of Conduct.\n\nSubmissions are due August 29, 2026.", "cfp_status": "published", "topics": [ "Open model development and community infrastructure", "Communication-efficient distributed training", "Asynchronous and local-update optimization", "Inter-datacenter training", "Internet-scale volunteer collaboration", "Federated learning at scale", "Heterogeneous devices and interconnects", "Fault tolerance, stragglers, and elasticity", "Parallelism strategies over low-bandwidth interconnects", "Security and robustness in trustless settings", "Incentives, governance, and economic models", "Benchmarks and evaluation" ], "important_dates": [ { "label": "Submission deadline", "date": "2026-08-29" }, { "label": "Reviewing deadline", "date": "2026-09-19" }, { "label": "Notification of acceptance", "date": "2026-09-29" }, { "label": "Camera-ready deadline", "date": "TBA" }, { "label": "Workshop day", "date": "2026-12-11" } ], "organizers": [ "Thalaiyasingam Ajanthan (Pluralis Research · ANU)", "Sameera Ramasinghe (Pluralis Research · University of Adelaide)", "Benjamin Thérien (Mila · Université de Montréal)", "Anastasia Koloskova (University of Zurich)", "Kaja Gruntkowska (KAUST)", "Eugene Belilovsky (Concordia · Mila)", "Aakanksha Chowdhery (Reflection AI · Stanford)", "Nicholas Lane (Cambridge · Flower Labs)" ], "speakers": [ "Peter Richtárik (KAUST)", "Nathan Lambert (Interconnects · formerly Ai2)", "Max Ryabinin (Together AI)", "Alexander Long (Pluralis Research)", "Virginia Smith (Carnegie Mellon University)" ], "host_url": "https://collaborative-open-decentralized-fomo.github.io/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/CODEC-FM", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/CODEC-FM", "location": "Sydney, Australia", "city": "Sydney", "workshop_date": "2026-12-10", "contact": "codec-fm-neurips-2026@googlegroups.com", "tracks": [ { "key": "CODEC-FM", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/CODEC-FM", "submission_dates_raw": "" } ], "group": "fm", "group_label": "Foundation Models & LLMs", "corpus": "NeurIPS Workshop on Collaborative, Open, and Decentralized Training of Foundation Models CODEC-FM 2026 CODEC-FM brings together the open-model and decentralized-training communities to address scaling foundation model training across distributed, heterogeneous compute — from inter-datacenter training to internet-scale volunteer collaboration. Open model development and community infrastructure Communication-efficient distributed training Asynchronous and local-update optimization Inter-datacenter training Internet-scale volunteer collaboration Federated learning at scale Heterogeneous devices and interconnects Fault tolerance, stragglers, and elasticity Parallelism strategies over low-bandwidth interconnects Security and robustness in trustless settings Incentives, governance, and economic models Benchmarks and evaluation About\n\nTraining large-scale foundation models today depends on massive, centralized GPU clusters that are inaccessible to most academic institutions, startups, and industries.\n\nThis concentration of compute creates high barriers to entry, centralizes AI innovation, and limits broader progress in developing and studying frontier-scale foundation models. Decentralization and resource pooling provide a way to enable large-scale runs without these barriers, by training models across geographically distributed and heterogeneous devices — from coordinated inter-datacenter settings to consumer devices collaborating over the internet.\n\nThis workshop brings together researchers and practitioners from the open-model and decentralized-training communities to address the core technical challenges of this paradigm, extending foundation model training beyond the confines of a single datacenter and supporting more scalable, collaborative open model development.\n\nTopics of interest\n- Open model development and community infrastructure\n- Communication-efficient distributed training\n- Asynchronous and local-update optimization\n- Inter-datacenter training\n- Internet-scale volunteer collaboration\n- Federated learning at scale\n- Heterogeneous devices and interconnects\n- Fault tolerance, stragglers, and elasticity\n- Parallelism strategies over low-bandwidth interconnects\n- Security and robustness in trustless settings\n- Incentives, governance, and economic models\n- Benchmarks and evaluation\n\nImportant Dates (All deadlines are 23:59 anywhere on Earth):\n- Submission deadline: August 29, 2026\n- Reviewing deadline: September 19, 2026\n- Notification of acceptance: September 29, 2026\n- Camera-ready deadline: TBA\n- Workshop day: December 11 or 12, 2026\n\nCall for Papers\n\nSubmission format\n- Long papers: up to 6 pages — eligible for oral presentation\n- Tiny papers: up to 2 pages — poster presentations only\n- Page limits exclude references and appendix\n- Use the official NeurIPS 2026 LaTeX style files (linked from the NeurIPS 2026 Call for Papers)\n- All submissions are non-archival\n- All authors are required to have a valid OpenReview profile\n- Submissions are managed through OpenReview\n\nWhat we accept\n- New research results\n- Position papers\n- Systems and infrastructure descriptions\n- Benchmark papers\n- Negative findings that clarify trade-offs\n\nTopics we accept\n\nWe encourage contributions addressing the core challenges of collaborative, open, and decentralized training, including but not limited to:\n- Open model development and community infrastructure: systems, architectures, algorithms, and governance mechanisms that enable large-scale collaborative open model development on geographically distributed compute.\n- Communication-efficient and asynchronous training: compression, scheduling, and optimization techniques that enable large-scale training across all parallelism axes over low-bandwidth and high-latency interconnects, while improving device utilization through asynchronous execution.\n- Heterogeneous and fault-tolerant systems: training algorithms and systems that support heterogeneous devices and interconnects in decentralized networks, maximize aggregate throughput, and remain robust to unreliable nodes, stragglers, and intermittent connectivity.\n- Security and robustness: methods for detecting malicious or faulty participants, ensuring robust convergence, and enabling secure collaboration in trustless or partially trusted decentralized training environments.\n\nReview process\n\nReviewing is double blind. Each submission will receive three reviews. Our program committee includes international experts with reviewing experience at top-tier conferences and recent workshops such as MCDC@ICLR'25 and FL@FM-NeurIPS'24. Organizers follow the NeurIPS conflict-of-interest policy. Six accepted papers will be selected for oral presentation, and all accepted papers will be presented in the poster session.\n\nAwards\n\nWe will present awards for the best paper, best student paper, and best tiny paper. Specific details will be communicated closer to the workshop date.\n\nCall for reviewers\n\nWe are recruiting reviewers from the broader community. If you work on distributed or decentralized training, federated learning, communication-efficient optimization, ML systems, or security for collaborative learning, we would be grateful for your help. To volunteer, fill out the reviewer sign-up form.\n\nDual submission\n\nBecause this workshop is non-archival, concurrent submission to other venues is permitted (subject to those venues' policies), and accepting a paper here does not preclude its later publication elsewhere. Accepted papers and the workshop schedule will be made publicly available on this website, and talks will be made public after the workshop.\n\nCode of conduct\n\nAll authors, reviewers, and attendees are expected to adhere to the NeurIPS Code of Conduct.\n\nSubmissions are due August 29, 2026." }, { "key": "ML4SpatialBio", "title": "NeurIPS Workshop on Machine Learning for Spatially Resolved High-dimensional Biology", "subtitle": "ML4SpatialBio 2026", "summary": "A one-day NeurIPS 2026 workshop establishing a dedicated venue for the intersection of machine learning and spatially resolved high-dimensional biology, bringing together ML researchers and experimental biologists to develop novel methods for spatial multiomics and tissue-scale data.", "cfp_full": "Machine Learning for Spatially Resolved High-dimensional Biology\n\nAdvancing ML for spatial biology: novel methods to integrate and analyze spatial multiomics and tissue-scale data. A one-day workshop bringing together ML researchers and experimental biologists.\n\nOverview\n\nSpatially resolved biology has become one of the most active frontiers in life sciences. Technologies for spatial transcriptomics, spatial proteomics and multiplex imaging now measure molecular activity while keeping the position of cells inside intact tissue. This makes it possible to study how cells communicate, organize, and change within their native microenvironment, with direct consequence for cancer research, immunology, neuroscience, developmental biology, and precision medicine.\n\nThese measures raise machine learning (ML) problems that do not fit cleanly into either single-cell or imaging pipelines. Tissue is a geometric object: cells sit in irregular spatial graphs, signals vary smoothly across space, and biological meaning depends on context at several scales at once, from subcellular structure to multicellular niche to whole organs. Spatial assays are also inherently multimodal, pairing molecular readouts with histology images, and they are growing toward atlas scale and whole-slide resolution. Geometry based learning, multimodal representation learning, generative modeling, and pretrained foundation models are well suited to these problems; however, they need adaptation and new theory to handle the noise, sparsity, and scale of high-dimensional spatial data.\n\nThis workshop establishes a dedicated venue at NeurIPS for the intersection of ML and spatially resolved high-dimensional biology. Our goal is to bring together ML researchers and experimental biologists around a shared set of methodological problems, to surface open challenges, and to begin building common benchmarks and evaluation standards for the field.\n\nTimeliness\n\nThe data landscape has shifted quickly. Platforms such as 10x Xenium, Visium HD, MERFISH, CosMx, and Phenocycler have moved spatial profiling from a handful of pilot studies into routine use, and public consortia are now releasing spatial atlases at large scale. The first spatial and tissue-aware foundation models have appeared in the past two years, and computational pathology has started to combine histology with molecular measurements. The methods, however, remain immature. There is still little agreement on how to represent tissue, how to evaluate spatial models, or how to transfer across platforms and tissues. This combination of abundant new data and unsettled methodology is precisely the moment when a focused NeurIPS workshop can shape the direction of a young field.\n\nTopics of Interest\n\nWe welcome contributions addressing open problems that call for novel ML methods rather than incremental applications, including:\n\n- Learning biologically meaningful spatial representations that hold across scales, from molecules to niches to organs\n- Integrating transcriptomic, proteomic, imaging, and clinical modalities into a single coherent model\n- Modeling cell-cell communication and tissue dynamics over time\n- Building interpretable and uncertainty-aware spatial models that biologists can trust\n- Designing benchmarks, datasets, and evaluation standards specific to spatial tasks\n- Developing novel ML and foundation models for spatial biology by integrating prior biological knowledge and inductive biases to improve interpretability and generalization across platforms, tissues, and species\n- Handling data sparsity, noise, batch effects, and limited annotations\n- Scaling learning to atlas-scale and whole-slide datasets\n\nSubmission format\n\nExtended abstracts up to 4 pages (references and optional appendix excluded). Submissions must use the official NeurIPS 2026 LaTeX template and be anonymized. Reviewing is double-blind, with three reviewers from the ML and biology communities. Submissions are managed via OpenReview. A contributed spotlight track reserves slots for early-career researchers, supporting diversity initiatives.\n\nImportant Dates:\n- Submission portal opens: August 2026\n- Submission deadline: August 29, 2026 (AoE)\n- Acceptance notifications: September 29, 2026 (AoE)\n- Workshop day: December 2026, Paris (TBD)", "cfp_status": "published", "topics": [ "Learning biologically meaningful spatial representations across scales (molecules to niches to organs)", "Integrating transcriptomic, proteomic, imaging, and clinical modalities into a single coherent model", "Modeling cell-cell communication and tissue dynamics over time", "Building interpretable and uncertainty-aware spatial models", "Designing benchmarks, datasets, and evaluation standards for spatial tasks", "Developing novel ML and foundation models integrating biological knowledge and inductive biases", "Handling data sparsity, noise, batch effects, and limited annotations", "Scaling learning to atlas-scale and whole-slide datasets" ], "important_dates": [ { "label": "Submission portal opens", "date": "August 2026" }, { "label": "Submission deadline", "date": "2026-08-29" }, { "label": "Acceptance notifications", "date": "2026-09-29" }, { "label": "Workshop day", "date": "December 2026 (TBD)" } ], "organizers": [ "Stefan Bonn (Professor & PI, University Medical Center Hamburg-Eppendorf (UKE), Germany)", "Robin Khatri (Postdoctoral Researcher & Junior Group Leader, UKE, Germany)", "Lucia Testa (Postdoctoral Researcher, UKE, Germany)", "Behnam Yousefi (Postdoctoral Researcher, UKE, Germany)" ], "speakers": [ "Julio Saez-Rodriguez (Head of Research, EMBL-EBI, Hinxton)", "Maria Brbić (Assistant Professor, EPFL, Lausanne)", "Pierre Bost (Junior Principal Investigator, Institut Curie, Paris)", "Martin Seifert (Senior Science & Technology Provider, 10x Genomics)" ], "host_url": "https://imsb-uke.github.io/ml4spatialbio-2026/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/ML4SpatialBio", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/ML4SpatialBio", "location": "Paris, France", "city": "Paris", "workshop_date": "", "contact": "ml4spatialbio.workshop@gmail.com", "tracks": [ { "key": "ML4SpatialBio", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/ML4SpatialBio", "submission_dates_raw": "Submission Deadline: Aug 29 2026 11:59PM UTC-0" } ], "group": "science", "group_label": "AI for Science", "corpus": "NeurIPS Workshop on Machine Learning for Spatially Resolved High-dimensional Biology ML4SpatialBio 2026 A one-day NeurIPS 2026 workshop establishing a dedicated venue for the intersection of machine learning and spatially resolved high-dimensional biology, bringing together ML researchers and experimental biologists to develop novel methods for spatial multiomics and tissue-scale data. Learning biologically meaningful spatial representations across scales (molecules to niches to organs) Integrating transcriptomic, proteomic, imaging, and clinical modalities into a single coherent model Modeling cell-cell communication and tissue dynamics over time Building interpretable and uncertainty-aware spatial models Designing benchmarks, datasets, and evaluation standards for spatial tasks Developing novel ML and foundation models integrating biological knowledge and inductive biases Handling data sparsity, noise, batch effects, and limited annotations Scaling learning to atlas-scale and whole-slide datasets Machine Learning for Spatially Resolved High-dimensional Biology\n\nAdvancing ML for spatial biology: novel methods to integrate and analyze spatial multiomics and tissue-scale data. A one-day workshop bringing together ML researchers and experimental biologists.\n\nOverview\n\nSpatially resolved biology has become one of the most active frontiers in life sciences. Technologies for spatial transcriptomics, spatial proteomics and multiplex imaging now measure molecular activity while keeping the position of cells inside intact tissue. This makes it possible to study how cells communicate, organize, and change within their native microenvironment, with direct consequence for cancer research, immunology, neuroscience, developmental biology, and precision medicine.\n\nThese measures raise machine learning (ML) problems that do not fit cleanly into either single-cell or imaging pipelines. Tissue is a geometric object: cells sit in irregular spatial graphs, signals vary smoothly across space, and biological meaning depends on context at several scales at once, from subcellular structure to multicellular niche to whole organs. Spatial assays are also inherently multimodal, pairing molecular readouts with histology images, and they are growing toward atlas scale and whole-slide resolution. Geometry based learning, multimodal representation learning, generative modeling, and pretrained foundation models are well suited to these problems; however, they need adaptation and new theory to handle the noise, sparsity, and scale of high-dimensional spatial data.\n\nThis workshop establishes a dedicated venue at NeurIPS for the intersection of ML and spatially resolved high-dimensional biology. Our goal is to bring together ML researchers and experimental biologists around a shared set of methodological problems, to surface open challenges, and to begin building common benchmarks and evaluation standards for the field.\n\nTimeliness\n\nThe data landscape has shifted quickly. Platforms such as 10x Xenium, Visium HD, MERFISH, CosMx, and Phenocycler have moved spatial profiling from a handful of pilot studies into routine use, and public consortia are now releasing spatial atlases at large scale. The first spatial and tissue-aware foundation models have appeared in the past two years, and computational pathology has started to combine histology with molecular measurements. The methods, however, remain immature. There is still little agreement on how to represent tissue, how to evaluate spatial models, or how to transfer across platforms and tissues. This combination of abundant new data and unsettled methodology is precisely the moment when a focused NeurIPS workshop can shape the direction of a young field.\n\nTopics of Interest\n\nWe welcome contributions addressing open problems that call for novel ML methods rather than incremental applications, including:\n\n- Learning biologically meaningful spatial representations that hold across scales, from molecules to niches to organs\n- Integrating transcriptomic, proteomic, imaging, and clinical modalities into a single coherent model\n- Modeling cell-cell communication and tissue dynamics over time\n- Building interpretable and uncertainty-aware spatial models that biologists can trust\n- Designing benchmarks, datasets, and evaluation standards specific to spatial tasks\n- Developing novel ML and foundation models for spatial biology by integrating prior biological knowledge and inductive biases to improve interpretability and generalization across platforms, tissues, and species\n- Handling data sparsity, noise, batch effects, and limited annotations\n- Scaling learning to atlas-scale and whole-slide datasets\n\nSubmission format\n\nExtended abstracts up to 4 pages (references and optional appendix excluded). Submissions must use the official NeurIPS 2026 LaTeX template and be anonymized. Reviewing is double-blind, with three reviewers from the ML and biology communities. Submissions are managed via OpenReview. A contributed spotlight track reserves slots for early-career researchers, supporting diversity initiatives.\n\nImportant Dates:\n- Submission portal opens: August 2026\n- Submission deadline: August 29, 2026 (AoE)\n- Acceptance notifications: September 29, 2026 (AoE)\n- Workshop day: December 2026, Paris (TBD)" }, { "key": "NEmo", "title": "Neuro-Symbolic Embodied Intelligence", "subtitle": "NEmo", "summary": "A workshop on embodied neuro-symbolic AI—integrating learned perception, reinforcement learning, large language models, planning, and structured symbolic knowledge in agents that act in physical or richly simulated environments.", "cfp_full": "Neuro-Symbolic Embodied Intelligence (NEmo) — Workshop at NeurIPS 2026\nEmbodied agents need neural perception, symbolic reasoning, and shared semantics to plan, learn, and act reliably.\n\nAbout the workshop\nNEmo focuses on embodied neuro-symbolic AI: the integration of learned perception, reinforcement learning, large language models, planning, and structured symbolic knowledge in agents that act in physical or richly simulated environments.\n\nRobotic and interactive agents must perceive changing worlds, reason over goals and constraints, execute long action sequences, and recover when assumptions fail. In this setting, hallucinated preconditions, brittle domain models, and ungrounded affordance knowledge become safety and reliability problems.\n\nThe workshop is designed for researchers in machine learning, knowledge representation, planning, robotics, human-robot interaction, semantic web, and trustworthy AI who are working on adjacent parts of the same problem.\n\nCore Challenges\nThe programme is organised around three technical challenges.\n- Long-Horizon Planning: Interfaces for learning, revising, and verifying symbolic action models while learned components handle perception and low-level control.\n- LLMs and Provenance: Reliable use of LLMs as domain-model elicitors, policy priors, symbolic constraint generators, and natural-language interfaces.\n- Shared Semantics: Reusable knowledge layers connecting environments, object affordances, robot capabilities, plans, and execution traces.\n\nTopics\n- Neuro-symbolic AI for embodied agents\n- LLMs for planning in atypical domains\n- Knowledge representation for planning\n- Ontological reasoning for embodied applications\n- Tacit planning knowledge in frontier models\n- Symbolic model elicitation for embodied reasoning\n- Perception and sensor grounding\n- Rule-based validation of symbolic domains\n- Neuro-symbolic reinforcement learning\n- Long-horizon planning and plan repair\n- Self-evolving agents\n- Ethical aspects of embodied AI systems\n\nWorkshop Format\nA one-day, in-person workshop of roughly 8.5 hours. The format combines keynotes, paper presentations, posters, demos, and structured breakout discussions. Organizers will introduce and moderate; keynotes will come from external speakers. The goal is to leave the day with a short public report covering open problems, shared assumptions, and candidate benchmarks or semantic layers. Demos are expected to be laptop-, video-, or browser-based unless additional equipment is secured.\n\nCall for Papers\nContributed work is non-archival and managed through OpenReview.\n\nLong Papers\n- Up to 8 pages, excluding references and appendices\n- Original work on embodied neuro-symbolic AI\n- Three reviews per submission\n\nShort Papers\n- Up to 4 pages, excluding references and appendices\n- Position papers, early results, benchmarks, demos\n- Eligible for lightning talks and posters\n\nSubmission and Review\nSubmit papers through OpenReview. Use the official NeurIPS 2026 LaTeX template. For workshop submissions, use the double-blind workshop option, dblblindworkshop. Submissions will be reviewed double-blind by the programme committee. Authors should anonymize manuscripts and supplementary material. NEmo is a non-proceedings workshop. Accepted papers will be published on the workshop website.\n\nDual-submission policy: The workshop will adopt a non-archival policy, welcoming ongoing and unpublished work, as well as papers under review or recently accepted at other venues. We discourage the submission of work previously published at major venues (e.g., NeurIPS, ICLR, ICML).\n\nImportant Dates\n- Submission deadline: September 10, 2026\n- Notifications sent: September 29, 2026\n- Camera-ready materials: October 2026\n- Workshop date: December 12, 2026 (Sydney)", "cfp_status": "published", "topics": [ "Neuro-symbolic AI for embodied agents", "LLMs for planning in atypical domains", "Knowledge representation for planning", "Ontological reasoning for embodied applications", "Tacit planning knowledge in frontier models", "Symbolic model elicitation for embodied reasoning", "Perception and sensor grounding", "Rule-based validation of symbolic domains", "Neuro-symbolic reinforcement learning", "Long-horizon planning and plan repair", "Self-evolving agents", "Ethical aspects of embodied AI systems" ], "important_dates": [ { "label": "Submission deadline", "date": "2026-09-10" }, { "label": "Notifications sent", "date": "2026-09-29" }, { "label": "Camera-ready materials", "date": "October 2026" }, { "label": "Workshop date", "date": "2026-12-12" } ], "organizers": [ "Elena Umili (Sapienza University of Rome)", "Emanuele Musumeci (Sapienza University of Rome)", "Vincenzo Suriani (Sapienza University of Rome)", "Daniel Dobriy (WU Vienna)", "Anna Sofia Lippolis (University of Bologna)" ], "speakers": [], "host_url": "https://nemo.semantic.review/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/NEmo", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/NEmo", "location": "Sydney", "city": "Sydney", "workshop_date": "2026-12-12", "contact": "pc@nemo.semantic.review", "tracks": [ { "key": "NEmo", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/NEmo", "submission_dates_raw": "Submission Start: Jul 28 2026 11:59PM UTC-0, Submission Deadline: Sep 10 2026 11:59PM UTC-0" } ], "group": "agents", "group_label": "Agents & Agentic Systems", "corpus": "Neuro-Symbolic Embodied Intelligence NEmo A workshop on embodied neuro-symbolic AI—integrating learned perception, reinforcement learning, large language models, planning, and structured symbolic knowledge in agents that act in physical or richly simulated environments. Neuro-symbolic AI for embodied agents LLMs for planning in atypical domains Knowledge representation for planning Ontological reasoning for embodied applications Tacit planning knowledge in frontier models Symbolic model elicitation for embodied reasoning Perception and sensor grounding Rule-based validation of symbolic domains Neuro-symbolic reinforcement learning Long-horizon planning and plan repair Self-evolving agents Ethical aspects of embodied AI systems Neuro-Symbolic Embodied Intelligence (NEmo) — Workshop at NeurIPS 2026\nEmbodied agents need neural perception, symbolic reasoning, and shared semantics to plan, learn, and act reliably.\n\nAbout the workshop\nNEmo focuses on embodied neuro-symbolic AI: the integration of learned perception, reinforcement learning, large language models, planning, and structured symbolic knowledge in agents that act in physical or richly simulated environments.\n\nRobotic and interactive agents must perceive changing worlds, reason over goals and constraints, execute long action sequences, and recover when assumptions fail. In this setting, hallucinated preconditions, brittle domain models, and ungrounded affordance knowledge become safety and reliability problems.\n\nThe workshop is designed for researchers in machine learning, knowledge representation, planning, robotics, human-robot interaction, semantic web, and trustworthy AI who are working on adjacent parts of the same problem.\n\nCore Challenges\nThe programme is organised around three technical challenges.\n- Long-Horizon Planning: Interfaces for learning, revising, and verifying symbolic action models while learned components handle perception and low-level control.\n- LLMs and Provenance: Reliable use of LLMs as domain-model elicitors, policy priors, symbolic constraint generators, and natural-language interfaces.\n- Shared Semantics: Reusable knowledge layers connecting environments, object affordances, robot capabilities, plans, and execution traces.\n\nTopics\n- Neuro-symbolic AI for embodied agents\n- LLMs for planning in atypical domains\n- Knowledge representation for planning\n- Ontological reasoning for embodied applications\n- Tacit planning knowledge in frontier models\n- Symbolic model elicitation for embodied reasoning\n- Perception and sensor grounding\n- Rule-based validation of symbolic domains\n- Neuro-symbolic reinforcement learning\n- Long-horizon planning and plan repair\n- Self-evolving agents\n- Ethical aspects of embodied AI systems\n\nWorkshop Format\nA one-day, in-person workshop of roughly 8.5 hours. The format combines keynotes, paper presentations, posters, demos, and structured breakout discussions. Organizers will introduce and moderate; keynotes will come from external speakers. The goal is to leave the day with a short public report covering open problems, shared assumptions, and candidate benchmarks or semantic layers. Demos are expected to be laptop-, video-, or browser-based unless additional equipment is secured.\n\nCall for Papers\nContributed work is non-archival and managed through OpenReview.\n\nLong Papers\n- Up to 8 pages, excluding references and appendices\n- Original work on embodied neuro-symbolic AI\n- Three reviews per submission\n\nShort Papers\n- Up to 4 pages, excluding references and appendices\n- Position papers, early results, benchmarks, demos\n- Eligible for lightning talks and posters\n\nSubmission and Review\nSubmit papers through OpenReview. Use the official NeurIPS 2026 LaTeX template. For workshop submissions, use the double-blind workshop option, dblblindworkshop. Submissions will be reviewed double-blind by the programme committee. Authors should anonymize manuscripts and supplementary material. NEmo is a non-proceedings workshop. Accepted papers will be published on the workshop website.\n\nDual-submission policy: The workshop will adopt a non-archival policy, welcoming ongoing and unpublished work, as well as papers under review or recently accepted at other venues. We discourage the submission of work previously published at major venues (e.g., NeurIPS, ICLR, ICML).\n\nImportant Dates\n- Submission deadline: September 10, 2026\n- Notifications sent: September 29, 2026\n- Camera-ready materials: October 2026\n- Workshop date: December 12, 2026 (Sydney)" }, { "key": "NewInML", "title": "New In Machine Learning (NewInML) Workshop @NeurIPS 2026", "subtitle": "NewInML workshop @ NeurIPS 2026", "summary": "A workshop designed for researchers who have not yet published at a top ML conference, pairing them with top researchers who review their work and share experience, with the goal of helping newcomers publish and contribute effectively to ML research.", "cfp_full": "New In ML at NeurIPS 2026\n\nNewInML will be held at NeurIPS 2026.\nParis, France | Date & Time: To Be Announced\n\nWe are here to help! The submissions are non archival. Feel free to submit elsewhere in the future! We are updating the website frequently. Stay tuned!\n\nOur Mission\n\nIs this your first time to a top conference? Have you ever wanted your own work recognized by this huge and active community? Do you encounter difficulties in polishing your ideas, experiments, paper writing, etc? Then, this session is exactly for you!\n\nThis year, we are hosting again the New in ML workshop at NeurIPS 2026. This workshop is intended for anyone who has not published a paper at a top conference yet (e.g. ICML, NeurIPS). We invited top researchers to review your work and share their experience with you. The best papers will receive awards and oral presentations. All accepted papers will be invited to present in our poster sessions!\n\nOur biggest goal is to help you publish papers at the next NeurIPS conference, and generally provide you with the guidance you need to contribute to ML research fully and effectively!\n\nCall for Papers\n\nWe invite submissions on all topics in machine learning. This workshop is especially designed for researchers who have not yet published at a top venue - we want to help you get there! The best papers will receive oral presentations and awards. All accepted papers will be presented as posters.\n\nSubmission Guidelines\nFormat: 2-8 pages (excluding references) using the NeurIPS 2026 workshop template.\nReview process: Double-blind via OpenReview. Submissions must be fully anonymized.\nNon-archival: Work may be submitted to or published at other venues concurrently.\nEligibility: Open to anyone who has not yet published at a top ML conference (NeurIPS, ICML, ICLR, etc.).\nAccepted posters: Authors are strongly encouraged to present a poster on-site at the workshop.\n\nImportant Dates\nAll deadlines are 11:59 PM Anywhere on Earth (AoE). Exact dates TBD.\nJuly 31, 2026 - Submission Portal Opens\nAugust 29, 2026 - Paper Submission Deadline\nSeptember 29, 2026 - Author Notification\nNovember 29, 2026 - Camera Ready Deadline\nTBD - Workshop Date, Paris, France\n\nNote: Excluding submissions that are not serious. For example: blank pages, topics not in machine learning and papers completely not following the format.", "cfp_status": "published", "topics": [ "All topics in machine learning (open to newcomers who have not yet published at a top ML venue)" ], "important_dates": [ { "label": "Submission Portal Opens", "date": "2026-07-31" }, { "label": "Paper Submission Deadline", "date": "2026-08-29" }, { "label": "Author Notification", "date": "2026-09-29" }, { "label": "Camera Ready Deadline", "date": "2026-11-29" } ], "organizers": [ "Arian Khorasani (University of Toronto & Vector Institute)", "Shashank Galla (Texas A&M University)", "Zarreen Reza (JACOBB Applied AI Center)", "Arth Singh (AIM Intelligence)", "Iñigo Parra (UC Berkeley)", "Veeraraju Elluru (SAAR-India)" ], "speakers": [ "Zhijing Jin (University of Toronto & Vector Institute, EuroSafeAI, Max Planck Institute for Intelligent Systems)", "Judah Goldfeder (Columbia University)", "Stella Biderman (EleutherAI)", "Niloofar Mireshghallah (humans&, Carnegie Mellon University)", "Natasha Jacques (University of Washington, Google DeepMind)", "Nathan Lambert (Allen Institute for AI)" ], "host_url": "https://newinml.github.io/NewInML2026NeurIPS/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/NewInML", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/NewInML", "location": "Paris, France", "city": "Paris", "workshop_date": "", "contact": "contactnewinml@gmail.com", "tracks": [ { "key": "NewInML", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/NewInML", "submission_dates_raw": "Submission Start: Aug 01 2026 12:00AM UTC-0, Submission Deadline: Aug 29 2026 08:59AM UTC-0" } ], "group": "society", "group_label": "Society, Applications & Community", "corpus": "New In Machine Learning (NewInML) Workshop @NeurIPS 2026 NewInML workshop @ NeurIPS 2026 A workshop designed for researchers who have not yet published at a top ML conference, pairing them with top researchers who review their work and share experience, with the goal of helping newcomers publish and contribute effectively to ML research. All topics in machine learning (open to newcomers who have not yet published at a top ML venue) New In ML at NeurIPS 2026\n\nNewInML will be held at NeurIPS 2026.\nParis, France | Date & Time: To Be Announced\n\nWe are here to help! The submissions are non archival. Feel free to submit elsewhere in the future! We are updating the website frequently. Stay tuned!\n\nOur Mission\n\nIs this your first time to a top conference? Have you ever wanted your own work recognized by this huge and active community? Do you encounter difficulties in polishing your ideas, experiments, paper writing, etc? Then, this session is exactly for you!\n\nThis year, we are hosting again the New in ML workshop at NeurIPS 2026. This workshop is intended for anyone who has not published a paper at a top conference yet (e.g. ICML, NeurIPS). We invited top researchers to review your work and share their experience with you. The best papers will receive awards and oral presentations. All accepted papers will be invited to present in our poster sessions!\n\nOur biggest goal is to help you publish papers at the next NeurIPS conference, and generally provide you with the guidance you need to contribute to ML research fully and effectively!\n\nCall for Papers\n\nWe invite submissions on all topics in machine learning. This workshop is especially designed for researchers who have not yet published at a top venue - we want to help you get there! The best papers will receive oral presentations and awards. All accepted papers will be presented as posters.\n\nSubmission Guidelines\nFormat: 2-8 pages (excluding references) using the NeurIPS 2026 workshop template.\nReview process: Double-blind via OpenReview. Submissions must be fully anonymized.\nNon-archival: Work may be submitted to or published at other venues concurrently.\nEligibility: Open to anyone who has not yet published at a top ML conference (NeurIPS, ICML, ICLR, etc.).\nAccepted posters: Authors are strongly encouraged to present a poster on-site at the workshop.\n\nImportant Dates\nAll deadlines are 11:59 PM Anywhere on Earth (AoE). Exact dates TBD.\nJuly 31, 2026 - Submission Portal Opens\nAugust 29, 2026 - Paper Submission Deadline\nSeptember 29, 2026 - Author Notification\nNovember 29, 2026 - Camera Ready Deadline\nTBD - Workshop Date, Paris, France\n\nNote: Excluding submissions that are not serious. For example: blank pages, topics not in machine learning and papers completely not following the format." }, { "key": "ODI", "title": "On-Device Intelligence: Foundation Models under Real-World Constraints", "subtitle": "ODI 2026", "summary": "ODI treats on-device intelligence as an interdisciplinary research problem spanning algorithms, systems, theory, hardware, and evaluation, focused on foundation models that jointly optimize efficiency, adaptation, execution, and reliability under real-world deployment constraints.", "cfp_full": "About the Workshop\n\nIntelligent systems are increasingly moving toward sustained interaction, online decision-making, and physical execution in the real world. In this transition, the traditional cloud-centric paradigm is becoming insufficient to meet practical requirements such as low latency, privacy preservation, safety and reliability, and robust operation under intermittent or limited connectivity.\n\nThe challenge. This workshop views on-device intelligence not merely as an extension of model compression or efficient deployment, but as a fundamentally interdisciplinary research problem spanning algorithms, systems, theory, hardware, and evaluation — where future on-device intelligence systems should jointly optimize efficiency, adaptation, execution, and reliability.\n\nOur goal. Recent advances in efficient foundation (multimodality) models, embodied intelligence, and hardware systems are progressing rapidly, yet often along parallel tracks with few shared abstractions or evaluation practices. We hope to create a shared venue for discussion and community building before these communities harden into separate problem formulations — helping clarify common challenges, identify novel research questions, and shape a coherent research agenda for on-device intelligence.\n\nJoin us for keynote talks, panel discussions, contributed oral presentations, and poster sessions!\n\nTopics & Call for Papers\n\nWe invite short paper submissions (5 pages, excluding references) using the NeurIPS 2026 LaTeX template, on all topics listed below but not limited to them.\n\n- Device-Native Foundation Model Design: Architectures, compression, quantization, distillation, and sparsity methods tailored for deployment under strict compute, memory, bandwidth, and energy constraints.\n- Efficient Adaptation, Inference and Reasoning under Real-World Constraints: Personalization, continual adaptation, and inference optimization strategies that operate reliably under the resource budgets of mobile, edge, and embedded systems.\n- Real-Time Multimodal and Embodied Intelligence: Low-latency perception, decision-making, and tool use for vision-language-action models, robotics, and autonomous systems operating directly on device.\n- Reliable, Safe, and Private Local Execution: Safety and privacy-preserving techniques for on-device models operating in offline or weak-connectivity settings.\n- Benchmarks and Evaluation for Interactive Real-World Deployment: Benchmarks and metrics that jointly assess performance, latency, energy, memory, safety, and reliability under realistic deployment conditions.\n\nThis workshop is non-archival and follows a double-blind review process. Outstanding submissions will be invited for a 15-minute oral presentation. All accepted papers will be presented during the poster session.\n\nImportant Dates (All deadlines are 23:59 Anywhere on Earth (AoE)):\n- Paper submission deadline: August 29, 2026\n- Review period: Aug 29 – Sep 19, 2026\n- Author notifications: September 29, 2026\n- Workshop day: December 11/12, 2026", "cfp_status": "published", "topics": [ "Device-Native Foundation Model Design (architectures, compression, quantization, distillation, sparsity under compute/memory/bandwidth/energy constraints)", "Efficient Adaptation, Inference and Reasoning under Real-World Constraints (personalization, continual adaptation, inference optimization for mobile/edge/embedded)", "Real-Time Multimodal and Embodied Intelligence (low-latency perception, decision-making, tool use for VLA models, robotics, autonomous systems on device)", "Reliable, Safe, and Private Local Execution (safety and privacy-preserving techniques for offline or weak-connectivity settings)", "Benchmarks and Evaluation for Interactive Real-World Deployment (metrics jointly assessing performance, latency, energy, memory, safety, reliability)" ], "important_dates": [ { "label": "Paper submission deadline", "date": "2026-08-29" }, { "label": "Review period", "date": "2026-08-29 to 2026-09-19" }, { "label": "Author notifications", "date": "2026-09-29" }, { "label": "Workshop day", "date": "2026-12-11" } ], "organizers": [ "Niao He (ETH Zurich)", "Bingcong Li (ETH Zurich)", "Shiwei Liu (ELLIS Institute Tübingen, MPI for Intelligent Systems, and Tübingen AI Center)", "Hao Ma (ETH Zurich)", "Michael Muehlebach (MPI for Intelligent Systems)", "Daniela Rus (MIT)", "Marko Zarić (MPI for Intelligent Systems)", "Melanie Zeilinger (ETH Zurich)" ], "speakers": [ "Yi Ma (The University of Hong Kong & UC Berkeley)", "Chelsea Finn (Stanford University & Physical Intelligence)", "Aishwarya Kamath (Google, Gemma 4 On-Device Integrations)", "Ted Xiao (Google DeepMind, Robotics & Vision-Language-Action Models)", "Dan Alistarh (ISTA, Efficient Machine Learning Systems)" ], "host_url": "https://odi2026.github.io/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/ODI", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/ODI", "location": "Sydney, Australia", "city": "Sydney", "workshop_date": "2026-12-11", "contact": "odi.neurips2026@gmail.com", "tracks": [ { "key": "ODI", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/ODI", "submission_dates_raw": "" } ], "group": "fm", "group_label": "Foundation Models & LLMs", "corpus": "On-Device Intelligence: Foundation Models under Real-World Constraints ODI 2026 ODI treats on-device intelligence as an interdisciplinary research problem spanning algorithms, systems, theory, hardware, and evaluation, focused on foundation models that jointly optimize efficiency, adaptation, execution, and reliability under real-world deployment constraints. Device-Native Foundation Model Design (architectures, compression, quantization, distillation, sparsity under compute/memory/bandwidth/energy constraints) Efficient Adaptation, Inference and Reasoning under Real-World Constraints (personalization, continual adaptation, inference optimization for mobile/edge/embedded) Real-Time Multimodal and Embodied Intelligence (low-latency perception, decision-making, tool use for VLA models, robotics, autonomous systems on device) Reliable, Safe, and Private Local Execution (safety and privacy-preserving techniques for offline or weak-connectivity settings) Benchmarks and Evaluation for Interactive Real-World Deployment (metrics jointly assessing performance, latency, energy, memory, safety, reliability) About the Workshop\n\nIntelligent systems are increasingly moving toward sustained interaction, online decision-making, and physical execution in the real world. In this transition, the traditional cloud-centric paradigm is becoming insufficient to meet practical requirements such as low latency, privacy preservation, safety and reliability, and robust operation under intermittent or limited connectivity.\n\nThe challenge. This workshop views on-device intelligence not merely as an extension of model compression or efficient deployment, but as a fundamentally interdisciplinary research problem spanning algorithms, systems, theory, hardware, and evaluation — where future on-device intelligence systems should jointly optimize efficiency, adaptation, execution, and reliability.\n\nOur goal. Recent advances in efficient foundation (multimodality) models, embodied intelligence, and hardware systems are progressing rapidly, yet often along parallel tracks with few shared abstractions or evaluation practices. We hope to create a shared venue for discussion and community building before these communities harden into separate problem formulations — helping clarify common challenges, identify novel research questions, and shape a coherent research agenda for on-device intelligence.\n\nJoin us for keynote talks, panel discussions, contributed oral presentations, and poster sessions!\n\nTopics & Call for Papers\n\nWe invite short paper submissions (5 pages, excluding references) using the NeurIPS 2026 LaTeX template, on all topics listed below but not limited to them.\n\n- Device-Native Foundation Model Design: Architectures, compression, quantization, distillation, and sparsity methods tailored for deployment under strict compute, memory, bandwidth, and energy constraints.\n- Efficient Adaptation, Inference and Reasoning under Real-World Constraints: Personalization, continual adaptation, and inference optimization strategies that operate reliably under the resource budgets of mobile, edge, and embedded systems.\n- Real-Time Multimodal and Embodied Intelligence: Low-latency perception, decision-making, and tool use for vision-language-action models, robotics, and autonomous systems operating directly on device.\n- Reliable, Safe, and Private Local Execution: Safety and privacy-preserving techniques for on-device models operating in offline or weak-connectivity settings.\n- Benchmarks and Evaluation for Interactive Real-World Deployment: Benchmarks and metrics that jointly assess performance, latency, energy, memory, safety, and reliability under realistic deployment conditions.\n\nThis workshop is non-archival and follows a double-blind review process. Outstanding submissions will be invited for a 15-minute oral presentation. All accepted papers will be presented during the poster session.\n\nImportant Dates (All deadlines are 23:59 Anywhere on Earth (AoE)):\n- Paper submission deadline: August 29, 2026\n- Review period: Aug 29 – Sep 19, 2026\n- Author notifications: September 29, 2026\n- Workshop day: December 11/12, 2026" }, { "key": "OPT", "title": "OPT 2026: Optimization for Machine Learning (NeurIPS 2026 Workshop)", "subtitle": "OPT 2026", "summary": "The 18th International OPT Workshop on Optimization for Machine Learning fosters discussion and dissemination of state-of-the-art research in optimization relevant to machine learning, with a 2026 focus on 'Can Anything Beat Adam? Frontier Optimizers.'", "cfp_full": "OPT 2026: Optimization for Machine Learning (NeurIPS 2026 Workshop)\n\nWe welcome you to participate in the 18th International OPT Workshop on Optimization for Machine Learning, to be held as a part of the NeurIPS 2026 conference. This year we particularly encourage (but not limit) submissions with a focus on \"Can Anything Beat Adam? Frontier Optimizers\".\n\nThe Workshop\nOptimization lies at the heart of many machine learning algorithms and continues to attract great interest across our community. This close connection between optimization and machine learning is the key motivation behind the OPT series of workshops. Now in its 18th edition, OPT aims to foster the discussion, discovery, and dissemination of state-of-the-art research in optimization relevant to machine learning.\n\nThe focus of OPT 2026 is \"Can Anything Beat Adam? Frontier Optimizers.\" Over the past decade, Adam and its variants have become the dominant methods for training machine learning models. Recently, however, new optimization algorithms, including non-diagonal and second-order methods such as Muon, K-FAC, and Shampoo, have shown promising empirical gains while remaining practical at scale. Their emergence raises fundamental questions about how data properties, stochasticity, batch size, model scale, and architecture should inform optimizer design. Under what conditions can these methods meaningfully outperform Adam? How can their practical gains be explained theoretically? And what are the limits of increasingly sophisticated optimizers? OPT 2026 will bring together researchers from theory and practice to advance our understanding and design of the next generation of optimization algorithms for machine learning.\n\nWe invite high-quality submissions for presentation as contributed talks, spotlights, or posters during the workshop.\n\nMain Topics\n- Adaptive Stochastic Methods\n- Higher-order methods and nonsmooth optimization\n- Adversarial Machine Learning approaches\n- Average-case Analysis of Optimization Algorithms\n- Benchmarking and empirical evaluation\n- Combinatorial optimization for machine learning\n- Deep learning optimization and large-scale model training\n- Federated learning\n- Frontier optimizers, preconditioning, and second-order methods\n- Game theory and min/max analysis\n- Nonconvex Optimization\n- Optimization software integration with deep learning tools\n- Parallel and Distributed Optimization\n- Privacy and Optimization\n- Scaling laws\n- Interface of Generalization and Optimization\n\nImportant Dates\n- Deadline for submission of papers: September 4, 2026 (AoE)\n- Notification of acceptance: September 29, 2026 (AoE)\n- Camera-ready papers upload to openreview: Nov 27, 2026 (OPT2026 style file required)\n- Workshop date: December 11 or 12, 2026\n\nSubmission Instructions\nPlatform: Submissions are made via OpenReview at https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/OPT. New profiles created without an institutional email address may require moderation, which can take up to two weeks.\nPage Limit (Initial Submission): Soft page limit of 5 pages, excluding references and appendices. Authors should use judgment; the main text should provide a coherent and self-contained explanation of the main contribution and its relevance to the workshop. Please do not submit an unchanged 9-page main-conference paper. Submissions with excessively long appendices, such as a 50-page proof, may also be considered a poor fit for the workshop.\nStyle and Format: Authors must use the OPT 2026 style file for submissions and camera-ready versions.\nAnonymization: The submission must be sufficiently anonymized for double-blind review.\nSupplementary Material: Appendices may contain additional results, proofs, or implementation details. However, submissions will primarily be evaluated based on the main text, and reviewers are not expected to read or verify all supplementary material.\nDual Submission Policy: The workshop will not accept submissions that have already been accepted for publication at another venue with archival proceedings. We do not accept dual submission to concurrent NeurIPS workshops; please choose the most suitable workshop for your submission — papers submitted to multiple workshops will be desk rejected. We tentatively accept submissions that are currently under review at NeurIPS, but require authors to retract their workshop submission if it is accepted at the main conference. Extended abstracts from other venues may be submitted provided that this is permitted by the other venue.\nCamera-Ready: 5-6 pages (without references and supplementary material); please use your own judgement (6 pages is a hard limit). Deadline November 27, 2026. Accepted submissions for which no de-anonymized camera-ready pdf has been uploaded by the deadline will be considered withdrawn.", "cfp_status": "published", "topics": [ "Adaptive Stochastic Methods", "Higher-order methods and nonsmooth optimization", "Adversarial Machine Learning approaches", "Average-case Analysis of Optimization Algorithms", "Benchmarking and empirical evaluation", "Combinatorial optimization for machine learning", "Deep learning optimization and large-scale model training", "Federated learning", "Frontier optimizers, preconditioning, and second-order methods", "Game theory and min/max analysis", "Nonconvex Optimization", "Optimization software integration with deep learning tools", "Parallel and Distributed Optimization", "Privacy and Optimization", "Scaling laws", "Interface of Generalization and Optimization" ], "important_dates": [ { "label": "Deadline for submission of papers", "date": "2026-09-04" }, { "label": "Notification of acceptance", "date": "2026-09-29" }, { "label": "Camera-ready papers upload", "date": "2026-11-27" }, { "label": "Workshop date", "date": "2026-12-11 or 2026-12-12" } ], "organizers": [ "Michael Crawshaw (Flatiron Institute)", "Frederik Kunstner (chair) (INRIA)", "Fred Roosta (University of Queensland)", "Courtney Paquette (McGill University / Google DeepMind)", "Sebastian Stich (CISPA Helmholtz Center)", "Chulhee \"Charlie\" Yun (KAIST)" ], "speakers": [ "Mateo Díaz (Johns Hopkins)", "Lénaïc Chizat (EPFL)", "Madeleine Udell (Stanford)", "Aaron Defazio (Meta)", "Jeremy Cohen (Flatiron Institute)" ], "host_url": "https://opt-ml.org/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/OPT", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/OPT", "location": "Sydney, Australia", "city": "Sydney", "workshop_date": "2026-12-11", "contact": "optmlworkshop@googlegroups.com", "tracks": [ { "key": "OPT", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/OPT", "submission_dates_raw": "" } ], "group": "theory", "group_label": "Theory & Methods", "corpus": "OPT 2026: Optimization for Machine Learning (NeurIPS 2026 Workshop) OPT 2026 The 18th International OPT Workshop on Optimization for Machine Learning fosters discussion and dissemination of state-of-the-art research in optimization relevant to machine learning, with a 2026 focus on 'Can Anything Beat Adam? Frontier Optimizers.' Adaptive Stochastic Methods Higher-order methods and nonsmooth optimization Adversarial Machine Learning approaches Average-case Analysis of Optimization Algorithms Benchmarking and empirical evaluation Combinatorial optimization for machine learning Deep learning optimization and large-scale model training Federated learning Frontier optimizers, preconditioning, and second-order methods Game theory and min/max analysis Nonconvex Optimization Optimization software integration with deep learning tools Parallel and Distributed Optimization Privacy and Optimization Scaling laws Interface of Generalization and Optimization OPT 2026: Optimization for Machine Learning (NeurIPS 2026 Workshop)\n\nWe welcome you to participate in the 18th International OPT Workshop on Optimization for Machine Learning, to be held as a part of the NeurIPS 2026 conference. This year we particularly encourage (but not limit) submissions with a focus on \"Can Anything Beat Adam? Frontier Optimizers\".\n\nThe Workshop\nOptimization lies at the heart of many machine learning algorithms and continues to attract great interest across our community. This close connection between optimization and machine learning is the key motivation behind the OPT series of workshops. Now in its 18th edition, OPT aims to foster the discussion, discovery, and dissemination of state-of-the-art research in optimization relevant to machine learning.\n\nThe focus of OPT 2026 is \"Can Anything Beat Adam? Frontier Optimizers.\" Over the past decade, Adam and its variants have become the dominant methods for training machine learning models. Recently, however, new optimization algorithms, including non-diagonal and second-order methods such as Muon, K-FAC, and Shampoo, have shown promising empirical gains while remaining practical at scale. Their emergence raises fundamental questions about how data properties, stochasticity, batch size, model scale, and architecture should inform optimizer design. Under what conditions can these methods meaningfully outperform Adam? How can their practical gains be explained theoretically? And what are the limits of increasingly sophisticated optimizers? OPT 2026 will bring together researchers from theory and practice to advance our understanding and design of the next generation of optimization algorithms for machine learning.\n\nWe invite high-quality submissions for presentation as contributed talks, spotlights, or posters during the workshop.\n\nMain Topics\n- Adaptive Stochastic Methods\n- Higher-order methods and nonsmooth optimization\n- Adversarial Machine Learning approaches\n- Average-case Analysis of Optimization Algorithms\n- Benchmarking and empirical evaluation\n- Combinatorial optimization for machine learning\n- Deep learning optimization and large-scale model training\n- Federated learning\n- Frontier optimizers, preconditioning, and second-order methods\n- Game theory and min/max analysis\n- Nonconvex Optimization\n- Optimization software integration with deep learning tools\n- Parallel and Distributed Optimization\n- Privacy and Optimization\n- Scaling laws\n- Interface of Generalization and Optimization\n\nImportant Dates\n- Deadline for submission of papers: September 4, 2026 (AoE)\n- Notification of acceptance: September 29, 2026 (AoE)\n- Camera-ready papers upload to openreview: Nov 27, 2026 (OPT2026 style file required)\n- Workshop date: December 11 or 12, 2026\n\nSubmission Instructions\nPlatform: Submissions are made via OpenReview at https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/OPT. New profiles created without an institutional email address may require moderation, which can take up to two weeks.\nPage Limit (Initial Submission): Soft page limit of 5 pages, excluding references and appendices. Authors should use judgment; the main text should provide a coherent and self-contained explanation of the main contribution and its relevance to the workshop. Please do not submit an unchanged 9-page main-conference paper. Submissions with excessively long appendices, such as a 50-page proof, may also be considered a poor fit for the workshop.\nStyle and Format: Authors must use the OPT 2026 style file for submissions and camera-ready versions.\nAnonymization: The submission must be sufficiently anonymized for double-blind review.\nSupplementary Material: Appendices may contain additional results, proofs, or implementation details. However, submissions will primarily be evaluated based on the main text, and reviewers are not expected to read or verify all supplementary material.\nDual Submission Policy: The workshop will not accept submissions that have already been accepted for publication at another venue with archival proceedings. We do not accept dual submission to concurrent NeurIPS workshops; please choose the most suitable workshop for your submission — papers submitted to multiple workshops will be desk rejected. We tentatively accept submissions that are currently under review at NeurIPS, but require authors to retract their workshop submission if it is accepted at the main conference. Extended abstracts from other venues may be submitted provided that this is permitted by the other venue.\nCamera-Ready: 5-6 pages (without references and supplementary material); please use your own judgement (6 pages is a hard limit). Deadline November 27, 2026. Accepted submissions for which no de-anonymized camera-ready pdf has been uploaded by the deadline will be considered withdrawn." }, { "key": "MPLR-FM", "title": "Privacy in the Era of Large Opaque Models: Theoretical, Legal, and Practical Perspectives", "subtitle": "PriLOM", "summary": "A NeurIPS 2026 workshop (PriLOM) examining memorization and privacy risks in opaque AI systems — foundation models, LLMs, and agentic systems — with emphasis on how these risks should be defined, measured, mitigated, and governed in deployment, bringing together researchers from privacy, ML, HCI, law, and agentic AI.", "cfp_full": "About the Workshop\nFoundation models, large language models, and agentic systems are rapidly reshaping the privacy landscape of machine learning. Across these paradigms, privacy risks are amplified by opacity: training data, post-training pipelines, alignment procedures, system components, and deployment contexts are often only partially visible to researchers, auditors, and users. These systems can expose sensitive information through memorization, retrieval, long-term memory, tool use, and cross-context information flow. At the same time, their opacity makes privacy assessment difficult. Together, these developments challenge existing definitions, benchmarks, mitigation strategies, governance frameworks, and accountability mechanisms.\n\nThis workshop will provide a timely forum for examining memorization and privacy risks in opaque AI systems, with emphasis on how these risks should be defined, measured, mitigated, and governed in deployment. It will bring together researchers from privacy, machine learning, HCI, law, and agentic AI around a central question: what should privacy mean when large opaque models are trained, adapted, deployed, and allowed to act on sensitive information?\n\nCall for Papers — Scope\nThe workshop seeks submissions addressing privacy challenges in foundation models and agentic systems. Key areas include evaluating and defining memorization, measuring tool-mediated data leakage, and understanding the privacy risks and failures that emerge from isolated or multi-step agent behavior.\n\nTopics of Interest\n- Privacy definitions for foundation models and agentic systems from contextual, legal, social, and formal angles\n- Memorization, leakage, and privacy attacks (extraction, inference, reconstruction, prompt injection, cross-session leakage, multi-agent risks)\n- Benchmarks, audits, and realistic evaluations of privacy risks and safeguards\n- Mitigations and formal protections (access control, unlearning, privacy-preserving tool use, differential privacy, verification)\n- Trade-offs, governance, and societal impacts regarding utility, safety, transparency, compliance, and downstream harms\n\nPosition Papers\nBeyond research papers, position papers presenting well-reasoned perspectives on challenges, opportunities, or future directions are welcomed. These should argue for viewpoints stimulating discussion and identifying open problems rather than reporting completed research. Position papers are assessed on argument quality and supporting evidence rather than reviewer agreement. Authors should clearly state their position in the title and abstract, present supporting evidence and reasoning, discuss credible alternative viewpoints, and conclude with concrete recommendations or research questions.\n\nSubmission Instructions\nFormat: Up to 4 pages, excluding references and appendices. Follow official NeurIPS 2026 formatting guidelines and use the NeurIPS 2026 LaTeX template; the paper checklist is not required. Original, unpublished work accepted; preprints on arXiv permitted. All submissions must be anonymized and submitted via OpenReview. All accepted papers require in-person presentation. This is a non-archival venue with no formal proceedings.\n\nDesk Rejection Criteria: non-compliance with template; failure to anonymize; violation of reciprocal reviewing agreement; breach of NeurIPS Code of Ethics; hallucinated content/citations; undocumented LLM use for important tasks.\n\nLLM Usage Policy: Document LLM/agent use for important, original, or non-standard tasks in appendix or main text. Basic spell-checking, grammar assistance, and code help need no documentation. Authors bear full responsibility for accuracy and originality. Agents and LLMs cannot be listed as authors.\n\nReview Policy: Submissions, reviews, and author responses remain visible only to assigned reviewers and organizers during review.\n\nImportant Dates\n- Submission deadline: August 29, 2026 (AoE)\n- Notification of decisions: September 29, 2026 (AoE)\n- Workshop: December 12 or 13, 2026, Paris, France", "cfp_status": "published", "topics": [ "Privacy definitions for foundation models and agentic systems (contextual, legal, social, formal)", "Memorization, leakage, and privacy attacks (extraction, inference, reconstruction, prompt injection, cross-session, multi-agent)", "Benchmarks, audits, and realistic evaluations of privacy risks and safeguards", "Mitigations and formal protections (access control, unlearning, privacy-preserving tool use, differential privacy, verification)", "Trade-offs, governance, and societal impacts (utility, safety, transparency, compliance, downstream harms)" ], "important_dates": [ { "label": "Submission Deadline", "date": "2026-08-29" }, { "label": "Notification of Decisions", "date": "2026-09-29" }, { "label": "Workshop", "date": "2026-12-12" } ], "organizers": [ "Sana Tonekaboni (Borealis AI and MIT)", "Lena Stempfle (MIT and Broad Institute)", "Linus Bleistein (EPFL AI Center)", "Maryam Molamohammadi (Mila - Quebec Artificial Intelligence Institute)", "Adam Dziedzic (CISPA Helmholtz Center for Information Security)", "Franziska Boenisch (CISPA Helmholtz Center for Information Security)" ], "speakers": [ "Aurélien Bellet (Professor at Inria)", "Florence G'sell (Professor at University of Lorraine)", "Reza Shokri (Google Zürich and Associate Professor at NUS)", "Carmela Troncoso (Associate Professor at EPFL)" ], "host_url": "https://neurips-workshop2026.github.io/foundation_model_agentic_privacy/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/MPLR-FM", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/MPLR-FM", "location": "Paris, France", "city": "Paris", "workshop_date": "2026-07-17", "contact": "adam.dziedzic@sprintml.com", "tracks": [ { "key": "MPLR-FM", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/MPLR-FM", "submission_dates_raw": "" } ], "group": "trust", "group_label": "Trustworthy & Safe AI", "corpus": "Privacy in the Era of Large Opaque Models: Theoretical, Legal, and Practical Perspectives PriLOM A NeurIPS 2026 workshop (PriLOM) examining memorization and privacy risks in opaque AI systems — foundation models, LLMs, and agentic systems — with emphasis on how these risks should be defined, measured, mitigated, and governed in deployment, bringing together researchers from privacy, ML, HCI, law, and agentic AI. Privacy definitions for foundation models and agentic systems (contextual, legal, social, formal) Memorization, leakage, and privacy attacks (extraction, inference, reconstruction, prompt injection, cross-session, multi-agent) Benchmarks, audits, and realistic evaluations of privacy risks and safeguards Mitigations and formal protections (access control, unlearning, privacy-preserving tool use, differential privacy, verification) Trade-offs, governance, and societal impacts (utility, safety, transparency, compliance, downstream harms) About the Workshop\nFoundation models, large language models, and agentic systems are rapidly reshaping the privacy landscape of machine learning. Across these paradigms, privacy risks are amplified by opacity: training data, post-training pipelines, alignment procedures, system components, and deployment contexts are often only partially visible to researchers, auditors, and users. These systems can expose sensitive information through memorization, retrieval, long-term memory, tool use, and cross-context information flow. At the same time, their opacity makes privacy assessment difficult. Together, these developments challenge existing definitions, benchmarks, mitigation strategies, governance frameworks, and accountability mechanisms.\n\nThis workshop will provide a timely forum for examining memorization and privacy risks in opaque AI systems, with emphasis on how these risks should be defined, measured, mitigated, and governed in deployment. It will bring together researchers from privacy, machine learning, HCI, law, and agentic AI around a central question: what should privacy mean when large opaque models are trained, adapted, deployed, and allowed to act on sensitive information?\n\nCall for Papers — Scope\nThe workshop seeks submissions addressing privacy challenges in foundation models and agentic systems. Key areas include evaluating and defining memorization, measuring tool-mediated data leakage, and understanding the privacy risks and failures that emerge from isolated or multi-step agent behavior.\n\nTopics of Interest\n- Privacy definitions for foundation models and agentic systems from contextual, legal, social, and formal angles\n- Memorization, leakage, and privacy attacks (extraction, inference, reconstruction, prompt injection, cross-session leakage, multi-agent risks)\n- Benchmarks, audits, and realistic evaluations of privacy risks and safeguards\n- Mitigations and formal protections (access control, unlearning, privacy-preserving tool use, differential privacy, verification)\n- Trade-offs, governance, and societal impacts regarding utility, safety, transparency, compliance, and downstream harms\n\nPosition Papers\nBeyond research papers, position papers presenting well-reasoned perspectives on challenges, opportunities, or future directions are welcomed. These should argue for viewpoints stimulating discussion and identifying open problems rather than reporting completed research. Position papers are assessed on argument quality and supporting evidence rather than reviewer agreement. Authors should clearly state their position in the title and abstract, present supporting evidence and reasoning, discuss credible alternative viewpoints, and conclude with concrete recommendations or research questions.\n\nSubmission Instructions\nFormat: Up to 4 pages, excluding references and appendices. Follow official NeurIPS 2026 formatting guidelines and use the NeurIPS 2026 LaTeX template; the paper checklist is not required. Original, unpublished work accepted; preprints on arXiv permitted. All submissions must be anonymized and submitted via OpenReview. All accepted papers require in-person presentation. This is a non-archival venue with no formal proceedings.\n\nDesk Rejection Criteria: non-compliance with template; failure to anonymize; violation of reciprocal reviewing agreement; breach of NeurIPS Code of Ethics; hallucinated content/citations; undocumented LLM use for important tasks.\n\nLLM Usage Policy: Document LLM/agent use for important, original, or non-standard tasks in appendix or main text. Basic spell-checking, grammar assistance, and code help need no documentation. Authors bear full responsibility for accuracy and originality. Agents and LLMs cannot be listed as authors.\n\nReview Policy: Submissions, reviews, and author responses remain visible only to assigned reviewers and organizers during review.\n\nImportant Dates\n- Submission deadline: August 29, 2026 (AoE)\n- Notification of decisions: September 29, 2026 (AoE)\n- Workshop: December 12 or 13, 2026, Paris, France" }, { "key": "PTA", "title": "PTA: From Pretrained Representations to Acting Agents", "subtitle": "NeurIPS 2026 Workshop PTA", "summary": "PTA focuses on the interface between pretraining and control: how pretrained representations learned from large-scale interaction data, videos, and multimodal corpora can be aligned with acting agents and transformed into reliable sequential decision-making behavior.", "cfp_full": "Sequential decision-making is undergoing a transition where agents increasingly rely on pretrained representations learned from large-scale interaction data, videos, and multimodal corpora, yet a central challenge remains: how can pretrained representations be aligned with acting agents and transformed into effective behavior?\n\nThe PTA workshop will focus on this emerging interface between pretraining and control. Pretrained models can encode rich semantic, temporal, causal, or structural information. Still, acting agents must transform that knowledge into reliable behavior, including choosing actions over time, adapting to new goals, reasoning under uncertainty, recovering from mistakes, and remaining robust under distribution shift.\n\nWe invite submissions that investigate how pretrained knowledge can become actionable for sequential decision-making. Beyond proposing new methods, we particularly encourage work that studies the principles, limitations, evaluation methodologies, and trade-offs underlying different approaches. By bringing together researchers from reinforcement learning, representation learning, robotics, planning, and foundation models, the workshop aims to facilitate discussion toward a deeper understanding of when and why pretrained representations enable effective decision making.\n\nTopics of Interest:\n- Actionable Representations: Successor representations, predictive state representations, world-model latents, representation geometry, abstraction, controllability, and connections between representation learning, planning, and generalization.\n- Test-Time Alignment: Test-time alignment with planning objectives, test-time search and optimization, online reasoning, retrieval, memory, verification, and tool use.\n- Transfer and Adaptation: Zero-shot and few-shot sequential decision-making, representation reuse across tasks and embodiments, multimodal and embodied representations for robotics.\n- Evaluation Methods: Evaluation protocols for robustness and transfer, benchmarks for long-horizon control, and diagnosing representation failures.\n- Failure Modes and Safety: Misalignment issues, hallucinated plans, unsafe exploration, and compounding errors in sequential decision-making.\n\nSubmission Instructions:\n- Follow NeurIPS 2026 paper guidelines (checklist not required).\n- Use OpenReview for double-blind review.\n- Two submission types: Full papers (up to 9 pages) with large-scale experiments; Short submissions (up to 4 pages) with proof-of-concept demonstrations.\n- Non-archival venue (concurrent submissions permitted).\n\nKey Dates:\n- Submission deadline: August 29, 2026, AoE\n- Author notification: September 29, 2026, AoE\n- Camera-ready deadline: November 25, 2026, AoE\n- Workshop date: December 11/12, 2026", "cfp_status": "published", "topics": [ "Actionable Representations (successor representations, predictive state representations, world-model latents, representation geometry, abstraction, controllability)", "Test-Time Alignment (planning objectives, test-time search and optimization, online reasoning, retrieval, memory, verification, and tool use)", "Transfer and Adaptation (zero-shot and few-shot sequential decision-making, representation reuse across tasks and embodiments, multimodal and embodied representations for robotics)", "Evaluation Methods (protocols for robustness and transfer, benchmarks for long-horizon control, diagnosing representation failures)", "Failure Modes and Safety (misalignment, hallucinated plans, unsafe exploration, compounding errors)" ], "important_dates": [ { "label": "Submission deadline", "date": "2026-08-29" }, { "label": "Author notification", "date": "2026-09-29" }, { "label": "Camera-ready deadline", "date": "2026-11-25" }, { "label": "Workshop date", "date": "2026-12-11" } ], "organizers": [ "Ping-Chun Hsieh (National Yang Ming Chiao Tung University)", "Kuang-Huei Lee (Google DeepMind)", "Bo Dai (Georgia Tech / Google DeepMind)", "Yen-Ling Kuo (University of Virginia)", "Georgia Chalvatzaki (TU Darmstadt and Hessian.AI)", "Karen Leung (University of Washington / NVIDIA Research)", "Co Yong (National Taiwan University)", "Claas Voelcker (University of Texas at Austin)" ], "speakers": [ "Sherry Yang (NYU Courant / Google DeepMind)", "Na (Lina) Li (Harvard University)", "Alexandre Proutiere (KTH Royal Institute of Technology)", "Aviral Kumar (CMU)", "Benjamin Eysenbach (Princeton)", "Adam White (University of Alberta / RL Core Technologies)", "Fabio Ramos (University of Sydney / NVIDIA)" ], "host_url": "https://ptaworkshop.github.io/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/PTA", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/PTA", "location": "Sydney, Australia", "city": "Sydney", "workshop_date": "2026-12-10", "contact": "pta.workshop2026@gmail.com", "tracks": [ { "key": "PTA", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/PTA", "submission_dates_raw": "" } ], "group": "agents", "group_label": "Agents & Agentic Systems", "corpus": "PTA: From Pretrained Representations to Acting Agents NeurIPS 2026 Workshop PTA PTA focuses on the interface between pretraining and control: how pretrained representations learned from large-scale interaction data, videos, and multimodal corpora can be aligned with acting agents and transformed into reliable sequential decision-making behavior. Actionable Representations (successor representations, predictive state representations, world-model latents, representation geometry, abstraction, controllability) Test-Time Alignment (planning objectives, test-time search and optimization, online reasoning, retrieval, memory, verification, and tool use) Transfer and Adaptation (zero-shot and few-shot sequential decision-making, representation reuse across tasks and embodiments, multimodal and embodied representations for robotics) Evaluation Methods (protocols for robustness and transfer, benchmarks for long-horizon control, diagnosing representation failures) Failure Modes and Safety (misalignment, hallucinated plans, unsafe exploration, compounding errors) Sequential decision-making is undergoing a transition where agents increasingly rely on pretrained representations learned from large-scale interaction data, videos, and multimodal corpora, yet a central challenge remains: how can pretrained representations be aligned with acting agents and transformed into effective behavior?\n\nThe PTA workshop will focus on this emerging interface between pretraining and control. Pretrained models can encode rich semantic, temporal, causal, or structural information. Still, acting agents must transform that knowledge into reliable behavior, including choosing actions over time, adapting to new goals, reasoning under uncertainty, recovering from mistakes, and remaining robust under distribution shift.\n\nWe invite submissions that investigate how pretrained knowledge can become actionable for sequential decision-making. Beyond proposing new methods, we particularly encourage work that studies the principles, limitations, evaluation methodologies, and trade-offs underlying different approaches. By bringing together researchers from reinforcement learning, representation learning, robotics, planning, and foundation models, the workshop aims to facilitate discussion toward a deeper understanding of when and why pretrained representations enable effective decision making.\n\nTopics of Interest:\n- Actionable Representations: Successor representations, predictive state representations, world-model latents, representation geometry, abstraction, controllability, and connections between representation learning, planning, and generalization.\n- Test-Time Alignment: Test-time alignment with planning objectives, test-time search and optimization, online reasoning, retrieval, memory, verification, and tool use.\n- Transfer and Adaptation: Zero-shot and few-shot sequential decision-making, representation reuse across tasks and embodiments, multimodal and embodied representations for robotics.\n- Evaluation Methods: Evaluation protocols for robustness and transfer, benchmarks for long-horizon control, and diagnosing representation failures.\n- Failure Modes and Safety: Misalignment issues, hallucinated plans, unsafe exploration, and compounding errors in sequential decision-making.\n\nSubmission Instructions:\n- Follow NeurIPS 2026 paper guidelines (checklist not required).\n- Use OpenReview for double-blind review.\n- Two submission types: Full papers (up to 9 pages) with large-scale experiments; Short submissions (up to 4 pages) with proof-of-concept demonstrations.\n- Non-archival venue (concurrent submissions permitted).\n\nKey Dates:\n- Submission deadline: August 29, 2026, AoE\n- Author notification: September 29, 2026, AoE\n- Camera-ready deadline: November 25, 2026, AoE\n- Workshop date: December 11/12, 2026" }, { "key": "QueerInAI", "title": "Queer in AI and {Dis}Ability in AI Workshop at NeurIPS 2026", "subtitle": "QueerInAI AbilityInAI 2026", "summary": "A joint affinity workshop and social for queer and/or disabled researchers, serving as a gathering space to build community and showcase work at the intersection of AI/ML with queerness and disability.", "cfp_full": "Queer in AI x {Dis}ability in AI @ NeurIPS 2026\n\nOur Call for Contributions is Live! Submit here. For more details, see the Call for Contributions section.\n\nMission\nQueer in AI's workshop and socials at NeurIPS 2026 aim to act as a gathering space for queer folks to build community and solidarity while enabling participants to learn about key issues and topics at the intersection of AI and queerness.\n\nPlanned Events\nNote that these events are currently tentative and it is possible our plans will change as we get closer to the conference. This website will have our most up-to-date schedule.\nSydney, Australia: Affinity poster session (in-person); Social (in-person); Full-day workshop (hybrid)\nParis, France: Affinity poster session (in-person); Social (in-person); Half-day workshop (hybrid)\nAtlanta, Georgia, USA: Affinity poster session (in-person); Social (in-person); Half-day workshop (hybrid)\nVirtual: Affinity poster session\n\nCall for Contributions\nSubmit here!\nNOTE: Queer in AI and {Dis}ability in AI are doing a joint workshop at NeurIPS this year. This means we will consider work that falls under either category for acceptance to our workshop:\n- Submissions from queer and/or disabled scholars\n- Submissions that explore the intersection of queerness + AI and/or disability + AI\nWe welcome submissions from queer and/or disabled researchers, as well as any and all work that explores the intersectionality of AI and ML with queer and/or disability issues, including but not limited to: addressing shortcomings in current state-of-the-art models, biases, and more queer-friendly research. Submissions can be in the form of research papers, extended abstracts, position papers, opinion pieces, surveys, or media pieces (art, music, videos, etc.). The workshop is non-archival, so full papers, work-in-progress papers, dual submissions, and works submitted or accepted elsewhere are all welcome!\n\nAccepted authors will have the opportunity to:\n- Have their work showcased by Queer in AI and {Dis}Ability in AI\n- Control how their name appears in public listings of accepted submissions\n- Present at the NeurIPS 2026 conference. We hope to be able to provide complimentary NeurIPS registrations to authors (stay tuned for updates!). The workshop is non-archival, so full papers, work-in-progress papers, dual submissions, and works submitted or accepted elsewhere are all welcome!\n\nGenerative AI Usage Policy\nWe require that authors use generative AI as a supportive tool at most. For the scope of our workshop, we do not wish to see generative AI used as a replacement for the process of creation. AI-generated submissions will be filtered out and not considered for inclusion in our workshop.\n\nImportant Dates (AoE = Anywhere on Earth)\nVisa-friendly submission deadline: August 10, 2026 AoE\nVisa-friendly notification deadline: August 20, 2026 AoE\nFinal submission deadline: September 03, 2026 AoE\nFinal notification deadline: September 17, 2026 AoE\n\nConference\nSydney, Australia: Dec 06-12\nParis, France: Dec 09-13\nAtlanta, Georgia, USA: Dec 09-13\n\nWe are so excited to review everyone's submissions + showcase our community's amazing work at NeurIPS 2026. Email us at queer-in-ai-neurips-2026@googlegroups.com with any questions, comments, or concerns.", "cfp_status": "published", "topics": [ "Submissions from queer and/or disabled scholars", "Intersection of queerness and AI", "Intersection of disability and AI", "Addressing shortcomings in current state-of-the-art models", "Biases in AI/ML", "Queer-friendly research", "Research papers, extended abstracts, position papers, opinion pieces, surveys, and media pieces (art, music, videos)" ], "important_dates": [ { "label": "Visa-friendly submission deadline", "date": "2026-08-10" }, { "label": "Visa-friendly notification deadline", "date": "2026-08-20" }, { "label": "Final submission deadline", "date": "2026-09-03" }, { "label": "Final notification deadline", "date": "2026-09-17" } ], "organizers": [ "Sharvani Jha (she/her)", "Alissa Valentine (she/they, University of Copenhagen)", "Kaiser Sun (they/them, Johns Hopkins University)", "Esmeralda S. Whitammer (she/they, University of Edinburgh)", "Ashwin S (she/they, TU Wien)", "Thibault Marette (he/him, KTH Royal Institute of Technology)", "Alex Markham (they/them, University of Copenhagen)", "Siyan Li (she/her, Columbia University)" ], "speakers": [], "host_url": "https://www.queerinai.com/neurips-2026", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/QueerInAI", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/QueerInAI", "location": "Sydney, Paris, Atlanta", "city": "Multiple", "workshop_date": "2026-12-05", "contact": "queer-in-ai-neurips-2026@googlegroups.com", "tracks": [ { "key": "QueerInAI", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/QueerInAI", "submission_dates_raw": "" } ], "group": "society", "group_label": "Society, Applications & Community", "corpus": "Queer in AI and {Dis}Ability in AI Workshop at NeurIPS 2026 QueerInAI AbilityInAI 2026 A joint affinity workshop and social for queer and/or disabled researchers, serving as a gathering space to build community and showcase work at the intersection of AI/ML with queerness and disability. Submissions from queer and/or disabled scholars Intersection of queerness and AI Intersection of disability and AI Addressing shortcomings in current state-of-the-art models Biases in AI/ML Queer-friendly research Research papers, extended abstracts, position papers, opinion pieces, surveys, and media pieces (art, music, videos) Queer in AI x {Dis}ability in AI @ NeurIPS 2026\n\nOur Call for Contributions is Live! Submit here. For more details, see the Call for Contributions section.\n\nMission\nQueer in AI's workshop and socials at NeurIPS 2026 aim to act as a gathering space for queer folks to build community and solidarity while enabling participants to learn about key issues and topics at the intersection of AI and queerness.\n\nPlanned Events\nNote that these events are currently tentative and it is possible our plans will change as we get closer to the conference. This website will have our most up-to-date schedule.\nSydney, Australia: Affinity poster session (in-person); Social (in-person); Full-day workshop (hybrid)\nParis, France: Affinity poster session (in-person); Social (in-person); Half-day workshop (hybrid)\nAtlanta, Georgia, USA: Affinity poster session (in-person); Social (in-person); Half-day workshop (hybrid)\nVirtual: Affinity poster session\n\nCall for Contributions\nSubmit here!\nNOTE: Queer in AI and {Dis}ability in AI are doing a joint workshop at NeurIPS this year. This means we will consider work that falls under either category for acceptance to our workshop:\n- Submissions from queer and/or disabled scholars\n- Submissions that explore the intersection of queerness + AI and/or disability + AI\nWe welcome submissions from queer and/or disabled researchers, as well as any and all work that explores the intersectionality of AI and ML with queer and/or disability issues, including but not limited to: addressing shortcomings in current state-of-the-art models, biases, and more queer-friendly research. Submissions can be in the form of research papers, extended abstracts, position papers, opinion pieces, surveys, or media pieces (art, music, videos, etc.). The workshop is non-archival, so full papers, work-in-progress papers, dual submissions, and works submitted or accepted elsewhere are all welcome!\n\nAccepted authors will have the opportunity to:\n- Have their work showcased by Queer in AI and {Dis}Ability in AI\n- Control how their name appears in public listings of accepted submissions\n- Present at the NeurIPS 2026 conference. We hope to be able to provide complimentary NeurIPS registrations to authors (stay tuned for updates!). The workshop is non-archival, so full papers, work-in-progress papers, dual submissions, and works submitted or accepted elsewhere are all welcome!\n\nGenerative AI Usage Policy\nWe require that authors use generative AI as a supportive tool at most. For the scope of our workshop, we do not wish to see generative AI used as a replacement for the process of creation. AI-generated submissions will be filtered out and not considered for inclusion in our workshop.\n\nImportant Dates (AoE = Anywhere on Earth)\nVisa-friendly submission deadline: August 10, 2026 AoE\nVisa-friendly notification deadline: August 20, 2026 AoE\nFinal submission deadline: September 03, 2026 AoE\nFinal notification deadline: September 17, 2026 AoE\n\nConference\nSydney, Australia: Dec 06-12\nParis, France: Dec 09-13\nAtlanta, Georgia, USA: Dec 09-13\n\nWe are so excited to review everyone's submissions + showcase our community's amazing work at NeurIPS 2026. Email us at queer-in-ai-neurips-2026@googlegroups.com with any questions, comments, or concerns." }, { "key": "RL4XS", "title": "Reinforcement Learning for Experimental Sciences Workshop", "subtitle": "RL4XS Workshop", "summary": "The RL4XS Workshop explores how reinforcement learning can enable adaptive experimentation and real-world scientific discovery while addressing the simulation-to-reality gap, bringing together machine learning researchers and experimental scientists.", "cfp_full": "Reinforcement Learning for Experimental Sciences\nBridging the Simulation-to-Reality Gap\nWorkshop at the Neural Information Processing Systems 2026 conference (Paris)\n\nWorkshop Motivation\n\nReinforcement learning has achieved major breakthroughs in simulated environments, yet transferring these advances to real experimental systems remains challenging. This workshop explores how RL can enable adaptive experimentation and real-world scientific discovery while addressing the simulation-to-reality gap.\n\nExperimental Systems as Sequential Decisions\nMany scientific experiments can naturally be formulated as sequential decision processes in which each experimental action influences future observations. Reinforcement learning provides a principled framework to optimize such adaptive decision loops.\n\nSimulation-to-Reality Gap\nDifferences between simulations and physical systems create major challenges for deploying reinforcement learning in laboratory and field experiments.\n\nReducing Experimental Cost and Time\nMany scientific experiments are costly, slow, or resource-constrained. Reinforcement learning methods can help allocate experimental effort more efficiently, accelerating discovery while reducing material, time, and financial costs.\n\nResearch Topics\n- RL for Experimental Platforms: Adaptive experiment design, autonomous laboratories, and sequential optimization in scientific experiments.\n- Bridging the Sim-to-Real Gap: Domain randomization, system identification, hybrid training pipelines, and robust policy transfer.\n- Efficient and Safe Learning: Sample-efficient reinforcement learning, safe exploration, uncertainty-aware decision making.\n- Field Experiments: Agricultural experimentation, ecological monitoring, and environmental sensing systems.\n- Human-in-the-Loop Science: Human–AI collaboration in experimentation and participatory research frameworks.\n- Benchmarks & Infrastructure: Simulation environments, digital twins, and benchmarking platforms for RL in experimental science.\n\nCall for Papers\n\nWe invite submissions describing reinforcement learning methods applied to real-world experimental systems, including research papers, system descriptions, benchmarks, and position papers. We will put efforts to include diverse participants. We encourage participation from both machine learning researchers and experimental scientists, practitioners, and interdisciplinary scientists.\n\nFormatting Instructions: We solicit workshop papers following the NeurIPS 2026 main conference paper template. The main text of a submitted paper is limited to nine content pages, including all figures and tables. Additional pages containing references, optional technical appendices and mandatory paper checklist do not count as content pages. The maximum size of submissions is 50 MB. While your submission can contain a supplement or appendix, please note that reviewers are not obliged to review supplementary material.\n\nReviews: The review process will be double-blind. All submissions must be anonymized and the leakage of any identification information is prohibited. All submissions will undergo peer review by the Program Committee. Each submission will receive at least three reviews, followed by a discussion phase among reviewers and organizers.\n\nPresentation: Accepted contributions will be presented as posters, with a subset selected for spotlight presentations.\n\nWhat can be submitted: Following the convention of most workshops, we welcome submissions of papers already published in journals or presented at non–machine-learning conferences or workshops. If a paper has appeared in a machine-learning venue, we will consider it only if it includes substantial extensions or new results.\n\nSubmission Timeline\n- Submission deadline: August 29, 2026\n- Notification: September 29, 2026\n- Camera-ready: November 29, 2026\n- Workshop: December 12 (Saturday), 2026", "cfp_status": "published", "topics": [ "RL for Experimental Platforms (adaptive experiment design, autonomous laboratories, sequential optimization)", "Bridging the Sim-to-Real Gap (domain randomization, system identification, hybrid training pipelines, robust policy transfer)", "Efficient and Safe Learning (sample-efficient RL, safe exploration, uncertainty-aware decision making)", "Field Experiments (agricultural experimentation, ecological monitoring, environmental sensing)", "Human-in-the-Loop Science (human–AI collaboration, participatory research frameworks)", "Benchmarks & Infrastructure (simulation environments, digital twins, benchmarking platforms)" ], "important_dates": [ { "label": "Submission deadline", "date": "2026-08-29" }, { "label": "Notification", "date": "2026-09-29" }, { "label": "Camera-ready", "date": "2026-11-29" }, { "label": "Workshop", "date": "2026-12-12" } ], "organizers": [ "Odalric-Ambrym Maillard (INRIA & University of Lille, France)", "Ronald Ortner (Technical University of Leoben, Austria)", "Audrey Durand (Université Laval, Canada)", "Sadegh Talebi (University of Copenhagen, Denmark)", "Samba Diaw (Higher Polytechnic School, Cheikh Anta DIOP University, Senegal)", "Tristan Fauvel (BaysicLabs, France)" ], "speakers": [ "Simon Hirländer (University of Salzburg)", "Sofia Villar (University of Cambridge)", "Alex Hernandez-Garcia (Université de Montréal & Mila, Canada)", "Ola Engkvist (AstraZeneca R&D & Chalmers University of Technology / University of Gothenburg)" ], "host_url": "https://rl-experimental-sciences-workshop.github.io/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/RL4XS", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/RL4XS", "location": "Paris, France", "city": "Paris", "workshop_date": "2026-12-12", "contact": "odalric.maillard@inria.fr", "tracks": [ { "key": "RL4XS", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/RL4XS", "submission_dates_raw": "" } ], "group": "science", "group_label": "AI for Science", "corpus": "Reinforcement Learning for Experimental Sciences Workshop RL4XS Workshop The RL4XS Workshop explores how reinforcement learning can enable adaptive experimentation and real-world scientific discovery while addressing the simulation-to-reality gap, bringing together machine learning researchers and experimental scientists. RL for Experimental Platforms (adaptive experiment design, autonomous laboratories, sequential optimization) Bridging the Sim-to-Real Gap (domain randomization, system identification, hybrid training pipelines, robust policy transfer) Efficient and Safe Learning (sample-efficient RL, safe exploration, uncertainty-aware decision making) Field Experiments (agricultural experimentation, ecological monitoring, environmental sensing) Human-in-the-Loop Science (human–AI collaboration, participatory research frameworks) Benchmarks & Infrastructure (simulation environments, digital twins, benchmarking platforms) Reinforcement Learning for Experimental Sciences\nBridging the Simulation-to-Reality Gap\nWorkshop at the Neural Information Processing Systems 2026 conference (Paris)\n\nWorkshop Motivation\n\nReinforcement learning has achieved major breakthroughs in simulated environments, yet transferring these advances to real experimental systems remains challenging. This workshop explores how RL can enable adaptive experimentation and real-world scientific discovery while addressing the simulation-to-reality gap.\n\nExperimental Systems as Sequential Decisions\nMany scientific experiments can naturally be formulated as sequential decision processes in which each experimental action influences future observations. Reinforcement learning provides a principled framework to optimize such adaptive decision loops.\n\nSimulation-to-Reality Gap\nDifferences between simulations and physical systems create major challenges for deploying reinforcement learning in laboratory and field experiments.\n\nReducing Experimental Cost and Time\nMany scientific experiments are costly, slow, or resource-constrained. Reinforcement learning methods can help allocate experimental effort more efficiently, accelerating discovery while reducing material, time, and financial costs.\n\nResearch Topics\n- RL for Experimental Platforms: Adaptive experiment design, autonomous laboratories, and sequential optimization in scientific experiments.\n- Bridging the Sim-to-Real Gap: Domain randomization, system identification, hybrid training pipelines, and robust policy transfer.\n- Efficient and Safe Learning: Sample-efficient reinforcement learning, safe exploration, uncertainty-aware decision making.\n- Field Experiments: Agricultural experimentation, ecological monitoring, and environmental sensing systems.\n- Human-in-the-Loop Science: Human–AI collaboration in experimentation and participatory research frameworks.\n- Benchmarks & Infrastructure: Simulation environments, digital twins, and benchmarking platforms for RL in experimental science.\n\nCall for Papers\n\nWe invite submissions describing reinforcement learning methods applied to real-world experimental systems, including research papers, system descriptions, benchmarks, and position papers. We will put efforts to include diverse participants. We encourage participation from both machine learning researchers and experimental scientists, practitioners, and interdisciplinary scientists.\n\nFormatting Instructions: We solicit workshop papers following the NeurIPS 2026 main conference paper template. The main text of a submitted paper is limited to nine content pages, including all figures and tables. Additional pages containing references, optional technical appendices and mandatory paper checklist do not count as content pages. The maximum size of submissions is 50 MB. While your submission can contain a supplement or appendix, please note that reviewers are not obliged to review supplementary material.\n\nReviews: The review process will be double-blind. All submissions must be anonymized and the leakage of any identification information is prohibited. All submissions will undergo peer review by the Program Committee. Each submission will receive at least three reviews, followed by a discussion phase among reviewers and organizers.\n\nPresentation: Accepted contributions will be presented as posters, with a subset selected for spotlight presentations.\n\nWhat can be submitted: Following the convention of most workshops, we welcome submissions of papers already published in journals or presented at non–machine-learning conferences or workshops. If a paper has appeared in a machine-learning venue, we will consider it only if it includes substantial extensions or new results.\n\nSubmission Timeline\n- Submission deadline: August 29, 2026\n- Notification: September 29, 2026\n- Camera-ready: November 29, 2026\n- Workshop: December 12 (Saturday), 2026" }, { "key": "RPS", "title": "Representations for the Physical Sciences Workshop @ NeurIPS 2026", "subtitle": "RPS Workshop 2026", "summary": "A NeurIPS 2026 workshop providing a tightly scoped forum for representation learning in physical systems, focused on self-supervision, transfer learning, sampling, and tokenization for scientific data and physical systems.", "cfp_full": "Representations for the Physical Sciences\n\nSelf-supervision, transfer learning, sampling, and tokenization for scientific data and physical systems.\n\nAbout the workshop\n\nRepresentations map observations into structured embeddings whose geometric structure reveals semantic content. In vision and language, deep learning models trained on large unlabeled datasets have produced representations that transfer broadly across tasks.\n\nAs AI progressively permeates scientific research, large-scale datasets together with high-throughput simulation and experimental pipelines, are making it possible to learn general-purpose scientific representations at unprecedented scale.\n\nThe scientific setting, however, challenges representation learning in fundamental ways. Data are heterogeneous and structured, and often come as unlabeled streams from a dynamical system. Moreover, the learned embeddings must respect conservation laws, geometry, and causal structure. Ultimately, a useful scientific representation should expose structures that scientists can act upon.\n\nThis workshop provides a tightly scoped forum for representation learning in physical systems. For this first edition, we welcome discussions and contributions on topics around Self-Supervision, Transfer Learning, Sampling, and Tokenization, which we believe constitute the most interesting open problems in this space.\n\nSelf-supervision (learning with no labels)\nScientific data are often abundant but unlabeled. Self-supervised learning offers a natural route to extracting structure from such data, but standard pretext tasks and augmentations can violate scientific meaning. What self-supervised objectives preserve physical constraints such as conservation laws? What semantic structures do SSL objectives discover when constrained by physical priors?\n\nTransfer (generalization)\nScientific models must reliably extrapolate into physically meaningful regimes beyond their training distribution. In the physical sciences, however, verifying a model's out-of-distribution prediction often requires massive computational effort or expensive wet-lab synthesis. This theme investigates the opportunities and limits of scientific transfer to ensure that learned embeddings remain falsifiable and actionable rather than just empirically successful.\n\nSampling (adaptive data generation)\nUnlike internet-scale text and image corpora, many scientific domains have access to simulators and experimental loops that can generate new data. Molecular dynamics, DFT, CFD, PDE solvers, and high-throughput experimental platforms make it possible to shape the training distribution itself. This makes adaptive simulation and closed-loop data generation a central opportunity for building better scientific representations.\n\nTokenization (physical modalities)\nThe success of foundation models relies heavily on discrete tokenization, but mapping physical sciences into discrete vocabularies remains a fundamental bottleneck. Physical data are inherently continuous, multi-scale, and often non-Euclidean. Naive grid-based patching or quantization can destroy geometric priors, symmetries, and sub-grid dynamics. This theme focuses on novel tokenization methods for physical systems.\n\nCall for papers\n\nWe invite interdisciplinary contributions from core machine learning and every area of AI for science.\n\nShort Papers (Non-archival)\nFour pages excluding references and appendices. Accepted work will be presented as posters, with selected submissions invited for contributed talks.\n\nShort Papers FAQs\n\nHow should I format my Short Paper?\nShort Papers are limited to four pages of main content; references do not count toward the limit. Appendices are unlimited, but reviewers are not obliged to read them, so keep the main paper self-contained. Unlike the NeurIPS main track, this workshop does not require the NeurIPS paper checklist. Download the workshop LaTeX template, which is based on the NeurIPS style. For styling details, consult the NeurIPS paper-formatting guidance.\n\nCan I include additional files, code, or data?\nYour submission must consist of a single PDF file including the main text, references, and, optionally, an appendix. Additional files are not allowed. You may link to properly anonymized code and/or data repositories.\n\nHow are Short Papers submitted and reviewed?\nShort Papers are submitted through OpenReview and reviewed double-blind, following the NeurIPS main-track approach. Concurrent submissions are allowed, but authors are responsible for checking the other venue's dual-submission policy.\n\nWhat are the requirements for Research Notes?\nResearch Notes have no page limit and will be submitted through a separate channel. Submission details will be announced soon.\n\nResearch Notes\nAccessible explanations of open problems, paradigms, datasets, or algorithms. New results are not required; clarity and usefulness are.\n\nImportant dates:\n- Call opens: 29 Jul 2026 · AoE\n- Target submission deadline: 29 Aug 2026 · AoE\n- Author notification: By 29 Sep 2026 · AoE", "cfp_status": "published", "topics": [ "Self-supervision for scientific data", "Transfer learning and generalization", "Sampling and adaptive data generation", "Tokenization for physical modalities", "Representation learning in physical systems", "AI for science" ], "important_dates": [ { "label": "Call opens", "date": "2026-07-29" }, { "label": "Target submission deadline", "date": "2026-08-29" }, { "label": "Author notification", "date": "2026-09-29" }, { "label": "Workshop", "date": "2026-12-12" } ], "organizers": [ "Pietro Novelli (Italian Institute of Technology)", "Massimiliano Pontil (Italian Institute of Technology · UCL)", "Florence d'Alché-Buc (Télécom Paris · IP Paris)", "Pierre Gentine (Columbia University)", "Kara Lamb (Columbia University)", "Mathias Niepert (University of Stuttgart)" ], "speakers": [ "Nils Thuerey (Technical University of Munich)", "Jean-Philippe Vert (Bioptimus)", "Mark Girolami (University of Cambridge)" ], "host_url": "https://representations-physical-sciences.github.io/workshop-2026/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/RPS", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/RPS", "location": "Paris, France", "city": "Paris", "workshop_date": "2026-12-12", "contact": "representations-physical-sciences@googlegroups.com", "tracks": [ { "key": "RPS", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/RPS", "submission_dates_raw": "" } ], "group": "science", "group_label": "AI for Science", "corpus": "Representations for the Physical Sciences Workshop @ NeurIPS 2026 RPS Workshop 2026 A NeurIPS 2026 workshop providing a tightly scoped forum for representation learning in physical systems, focused on self-supervision, transfer learning, sampling, and tokenization for scientific data and physical systems. Self-supervision for scientific data Transfer learning and generalization Sampling and adaptive data generation Tokenization for physical modalities Representation learning in physical systems AI for science Representations for the Physical Sciences\n\nSelf-supervision, transfer learning, sampling, and tokenization for scientific data and physical systems.\n\nAbout the workshop\n\nRepresentations map observations into structured embeddings whose geometric structure reveals semantic content. In vision and language, deep learning models trained on large unlabeled datasets have produced representations that transfer broadly across tasks.\n\nAs AI progressively permeates scientific research, large-scale datasets together with high-throughput simulation and experimental pipelines, are making it possible to learn general-purpose scientific representations at unprecedented scale.\n\nThe scientific setting, however, challenges representation learning in fundamental ways. Data are heterogeneous and structured, and often come as unlabeled streams from a dynamical system. Moreover, the learned embeddings must respect conservation laws, geometry, and causal structure. Ultimately, a useful scientific representation should expose structures that scientists can act upon.\n\nThis workshop provides a tightly scoped forum for representation learning in physical systems. For this first edition, we welcome discussions and contributions on topics around Self-Supervision, Transfer Learning, Sampling, and Tokenization, which we believe constitute the most interesting open problems in this space.\n\nSelf-supervision (learning with no labels)\nScientific data are often abundant but unlabeled. Self-supervised learning offers a natural route to extracting structure from such data, but standard pretext tasks and augmentations can violate scientific meaning. What self-supervised objectives preserve physical constraints such as conservation laws? What semantic structures do SSL objectives discover when constrained by physical priors?\n\nTransfer (generalization)\nScientific models must reliably extrapolate into physically meaningful regimes beyond their training distribution. In the physical sciences, however, verifying a model's out-of-distribution prediction often requires massive computational effort or expensive wet-lab synthesis. This theme investigates the opportunities and limits of scientific transfer to ensure that learned embeddings remain falsifiable and actionable rather than just empirically successful.\n\nSampling (adaptive data generation)\nUnlike internet-scale text and image corpora, many scientific domains have access to simulators and experimental loops that can generate new data. Molecular dynamics, DFT, CFD, PDE solvers, and high-throughput experimental platforms make it possible to shape the training distribution itself. This makes adaptive simulation and closed-loop data generation a central opportunity for building better scientific representations.\n\nTokenization (physical modalities)\nThe success of foundation models relies heavily on discrete tokenization, but mapping physical sciences into discrete vocabularies remains a fundamental bottleneck. Physical data are inherently continuous, multi-scale, and often non-Euclidean. Naive grid-based patching or quantization can destroy geometric priors, symmetries, and sub-grid dynamics. This theme focuses on novel tokenization methods for physical systems.\n\nCall for papers\n\nWe invite interdisciplinary contributions from core machine learning and every area of AI for science.\n\nShort Papers (Non-archival)\nFour pages excluding references and appendices. Accepted work will be presented as posters, with selected submissions invited for contributed talks.\n\nShort Papers FAQs\n\nHow should I format my Short Paper?\nShort Papers are limited to four pages of main content; references do not count toward the limit. Appendices are unlimited, but reviewers are not obliged to read them, so keep the main paper self-contained. Unlike the NeurIPS main track, this workshop does not require the NeurIPS paper checklist. Download the workshop LaTeX template, which is based on the NeurIPS style. For styling details, consult the NeurIPS paper-formatting guidance.\n\nCan I include additional files, code, or data?\nYour submission must consist of a single PDF file including the main text, references, and, optionally, an appendix. Additional files are not allowed. You may link to properly anonymized code and/or data repositories.\n\nHow are Short Papers submitted and reviewed?\nShort Papers are submitted through OpenReview and reviewed double-blind, following the NeurIPS main-track approach. Concurrent submissions are allowed, but authors are responsible for checking the other venue's dual-submission policy.\n\nWhat are the requirements for Research Notes?\nResearch Notes have no page limit and will be submitted through a separate channel. Submission details will be announced soon.\n\nResearch Notes\nAccessible explanations of open problems, paradigms, datasets, or algorithms. New results are not required; clarity and usefulness are.\n\nImportant dates:\n- Call opens: 29 Jul 2026 · AoE\n- Target submission deadline: 29 Aug 2026 · AoE\n- Author notification: By 29 Sep 2026 · AoE" }, { "key": "RCMLR", "title": "Responsible Communication of Machine Learning Research in Biomedicine", "subtitle": "RCMLR 2026", "summary": "A workshop that bridges the translation gap between machine learning researchers and the domain scientists, clinicians, and policymakers who must interpret and act on ML findings in biomedicine, treating structured interdisciplinary dialogue as its method.", "cfp_full": "Responsible Communication of Machine Learning Research in Biomedicine\nNeurIPS Workshop\nSydney, Australia · December 2026\n\nCall for papers\nThe 'Responsible Communication of Machine Learning Research in Biomedicine' workshop will bridge the translation gap between machine learning researchers and the domain scientists, clinicians and policymakers who must interpret and act on their findings.\n\nA persistent gap has emerged between what machine learning (ML) systems can deliver in biomedical contexts and how their capabilities are communicated to those who use or regulate them. In early discovery, high-profile systems such as AlphaFold and agentic research platforms such as AI Scientists have increasingly accelerated the pace at which unchecked capability claims enter public and policy discourse. In clinical deployment, decision-making tools are expanding rapidly but frameworks for communicating their limitations remain underdeveloped, illustrated by well-documented cases of oncology decision-support tools overstating their clinical capabilities, as well as more recent findings that LLMs achieving near-perfect medical benchmark scores fail to improve clinical decision-making with real patients. Across the pipeline, this gap drives hype, misuse, misinterpretation and poorly informed governance, with direct consequences for user trust, funding priorities and the effective adoption of advances in ML.\n\nIn early discovery, frameworks to address this challenge are largely absent; in clinical settings, reporting standards such as TRIPOD+AI and TRIPOD-LLM represent important steps but adherence remains low. A deeper contributing challenge is that the technical conventions and vocabulary that make findings legible within ML do not translate cleanly across the diverse stakeholders involved: researchers, clinicians, policymakers and those responsible for communicating advances more broadly frequently lack a common language and are left uncertain about what ML systems can and cannot do and unable to evaluate their claims. Because the challenges arise in the translation between communities, reporting standards alone or solutions developed by a single community in isolation cannot feasibly close the gap that unintended miscommunication creates.\n\nIn response, this workshop treats structured interdisciplinary dialogue as the method. Initiated from within the ML research community, it brings those who produce ML findings into direct exchange with the domain scientists, clinicians, policymakers, journalists and science communicators who must interpret and act on them. Previous NeurIPS, ICML and ICLR workshops have advanced related themes primarily from the ML perspective, including explainability, interpretability and responsible AI. This workshop builds on that foundation by shifting focus from how findings are communicated within ML to how they translate across the biomedical landscape and those who shape it. Grounded in real-world case studies, the workshop is designed to surface opportunities for evidence-based communication approaches when moving from problem to practice, with interdisciplinary exchange at its core.\n\nImportant dates\nCall for papers: Now open - submission via OpenReview\nSubmission deadline: 29 August 2026 (AOE)\nAuthor notifications: 29 September 2026 (AOE)\nWorkshop date: 11 or 12 December 2026 (TBC)\n\nComing soon: Speaker lineup and agenda will be shared here in advance of the workshop.\n\nQuestions? Contact the organizers: translating.ml.research@gmail.com", "cfp_status": "published", "topics": [ "Bridging the translation gap between ML researchers and biomedical stakeholders", "Communicating capabilities and limitations of ML systems in biomedicine", "Responsible communication in early biomedical discovery (e.g., AlphaFold, AI Scientists)", "Communication of clinical decision-support tool limitations", "Reporting standards such as TRIPOD+AI and TRIPOD-LLM", "Evidence-based communication approaches grounded in real-world case studies", "Interdisciplinary dialogue across researchers, clinicians, policymakers, journalists, and science communicators" ], "important_dates": [ { "label": "Submission deadline", "date": "2026-08-29" }, { "label": "Author notifications", "date": "2026-09-29" }, { "label": "Workshop date", "date": "2026-12-11" } ], "organizers": [ "Siobhan Sanford (GSK AIML)", "Julia Meister (GSK AIML)", "Qiyao Wei (Cambridge University)", "Galvin Khara (GSK AIML)", "Nikhil Kurian (Adelaide University)", "Ben Glocker (Imperial College)", "Jessica Schrouff (GSK AIML)" ], "speakers": [], "host_url": "https://translatingmlresearch.github.io/RCMLR/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/RCMLR", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/RCMLR", "location": "Sydney, Australia", "city": "Sydney", "workshop_date": "2026-12-10", "contact": "translating.ml.research@gmail.com", "tracks": [ { "key": "RCMLR", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/RCMLR", "submission_dates_raw": "" } ], "group": "health", "group_label": "Health & Medicine", "corpus": "Responsible Communication of Machine Learning Research in Biomedicine RCMLR 2026 A workshop that bridges the translation gap between machine learning researchers and the domain scientists, clinicians, and policymakers who must interpret and act on ML findings in biomedicine, treating structured interdisciplinary dialogue as its method. Bridging the translation gap between ML researchers and biomedical stakeholders Communicating capabilities and limitations of ML systems in biomedicine Responsible communication in early biomedical discovery (e.g., AlphaFold, AI Scientists) Communication of clinical decision-support tool limitations Reporting standards such as TRIPOD+AI and TRIPOD-LLM Evidence-based communication approaches grounded in real-world case studies Interdisciplinary dialogue across researchers, clinicians, policymakers, journalists, and science communicators Responsible Communication of Machine Learning Research in Biomedicine\nNeurIPS Workshop\nSydney, Australia · December 2026\n\nCall for papers\nThe 'Responsible Communication of Machine Learning Research in Biomedicine' workshop will bridge the translation gap between machine learning researchers and the domain scientists, clinicians and policymakers who must interpret and act on their findings.\n\nA persistent gap has emerged between what machine learning (ML) systems can deliver in biomedical contexts and how their capabilities are communicated to those who use or regulate them. In early discovery, high-profile systems such as AlphaFold and agentic research platforms such as AI Scientists have increasingly accelerated the pace at which unchecked capability claims enter public and policy discourse. In clinical deployment, decision-making tools are expanding rapidly but frameworks for communicating their limitations remain underdeveloped, illustrated by well-documented cases of oncology decision-support tools overstating their clinical capabilities, as well as more recent findings that LLMs achieving near-perfect medical benchmark scores fail to improve clinical decision-making with real patients. Across the pipeline, this gap drives hype, misuse, misinterpretation and poorly informed governance, with direct consequences for user trust, funding priorities and the effective adoption of advances in ML.\n\nIn early discovery, frameworks to address this challenge are largely absent; in clinical settings, reporting standards such as TRIPOD+AI and TRIPOD-LLM represent important steps but adherence remains low. A deeper contributing challenge is that the technical conventions and vocabulary that make findings legible within ML do not translate cleanly across the diverse stakeholders involved: researchers, clinicians, policymakers and those responsible for communicating advances more broadly frequently lack a common language and are left uncertain about what ML systems can and cannot do and unable to evaluate their claims. Because the challenges arise in the translation between communities, reporting standards alone or solutions developed by a single community in isolation cannot feasibly close the gap that unintended miscommunication creates.\n\nIn response, this workshop treats structured interdisciplinary dialogue as the method. Initiated from within the ML research community, it brings those who produce ML findings into direct exchange with the domain scientists, clinicians, policymakers, journalists and science communicators who must interpret and act on them. Previous NeurIPS, ICML and ICLR workshops have advanced related themes primarily from the ML perspective, including explainability, interpretability and responsible AI. This workshop builds on that foundation by shifting focus from how findings are communicated within ML to how they translate across the biomedical landscape and those who shape it. Grounded in real-world case studies, the workshop is designed to surface opportunities for evidence-based communication approaches when moving from problem to practice, with interdisciplinary exchange at its core.\n\nImportant dates\nCall for papers: Now open - submission via OpenReview\nSubmission deadline: 29 August 2026 (AOE)\nAuthor notifications: 29 September 2026 (AOE)\nWorkshop date: 11 or 12 December 2026 (TBC)\n\nComing soon: Speaker lineup and agenda will be shared here in advance of the workshop.\n\nQuestions? Contact the organizers: translating.ml.research@gmail.com" }, { "key": "RoboPAD", "title": "RoboPAD: Post-Training Adaptation of Robot Foundation Models", "subtitle": "NeurIPS 2026 RoboPAD", "summary": "A NeurIPS 2026 workshop consolidating post-training - adaptation, correction, evaluation, and safe improvement of pretrained robot foundation models - as a shared research agenda, bringing together robot learning, reinforcement learning, foundation models, world models, and embodied AI.", "cfp_full": "About the Workshop\n\nPretrained robot foundation models are increasingly capable, yet real-world deployment still requires adaptation to new environments, embodiments, tasks, and failures. RoboPAD focuses on post-training: how to adapt, correct, evaluate, and safely improve these models after pretraining. Post-training is the algorithmic stage after pretraining: the closed-loop process of adapting, correcting, evaluating, and improving robot foundation models under real-world constraints.\n\nWe bring together researchers in robot learning, reinforcement learning, foundation models, world models, and embodied AI to consolidate post-training as a shared research agenda.\n\nTopics\n- Human Feedback / Intervention - Feedback- and intervention-driven adaptation: human corrections, teleoperation, preference learning.\n- Policy Optimization / Specialization - Policy optimization after pretraining: RL fine-tuning, action decoders, fast specialization.\n- World Model / Simulation - World-model and simulation-driven adaptation: predictive models, simulation data, closed-loop refinement.\n- Reasoning & Memory - Reasoning, memory, and agentic adaptation: deliberative reasoning, task decomposition, recovery.\n- Cross-Embodiment Adaptation - Cross-embodiment and whole-body adaptation: humanoids, dexterous hands, sim-to-real transfer.\n- Evaluation & Safety - Evaluation, safety, and robustness: benchmarks, reproducible protocols, constraint-respecting fine-tuning.\n\nCall for Papers\n\nWe welcome work in progress, under-review work, and unpublished or preprint work substantially reframed as open-problem, benchmark, or perspective contributions. Archival published papers are not eligible. Dual submissions must comply with other venues' policies.\n\nSubmission Tracks\n- Short papers (up to 4 pages) - preliminary findings, positions, benchmarks, infrastructure reports, and negative results.\n- Long papers (up to 9 pages) - mature contributions with fuller technical development, experiments, or analysis.\nPage limits exclude references and appendices.\n\nReview & status\nDouble-blind via OpenReview, with three reviewers per paper; non-archival (accepted papers posted on OpenReview unless authors opt out); posters + spotlights + Best Paper. At least one author must present in person.\n\nImportant Dates\n- Submission: Aug 29, 2026 (Anywhere on Earth / AoE)\n- Notification: Sep 29, 2026 (Anywhere on Earth / AoE)\n- Workshop: Dec 12-13, 2026, Paris, France\n\nAssociated Challenge - RoboWorld Challenge 2026\nRoboWorld 2026 is an independently organized challenge associated with the RoboPAD Workshop at NeurIPS 2026. It focuses on embodied world modeling for robotics and autonomous driving across multiple tracks. Challenge logistics, registration, and evaluation are managed by the RoboWorld organizing team; participation is separate from RoboPAD paper submission.", "cfp_status": "published", "topics": [ "Human Feedback / Intervention (human corrections, teleoperation, preference learning)", "Policy Optimization / Specialization (RL fine-tuning, action decoders, fast specialization)", "World Model / Simulation (predictive models, simulation data, closed-loop refinement)", "Reasoning & Memory (deliberative reasoning, task decomposition, recovery)", "Cross-Embodiment Adaptation (humanoids, dexterous hands, sim-to-real transfer)", "Evaluation & Safety (benchmarks, reproducible protocols, constraint-respecting fine-tuning)" ], "important_dates": [ { "label": "Submission", "date": "2026-08-29" }, { "label": "Notification", "date": "2026-09-29" }, { "label": "Workshop", "date": "December 12-13, 2026" } ], "organizers": [ "Shijie Li (A*STAR)", "Shizhe Chen (Inria Paris)", "Ziwei Wang (NTU)", "Jiafei Duan (University of Washington)", "Sihao Lin (Adelaide University)", "Changliu Liu (Carnegie Mellon University)", "Gerhard Neumann (KIT)" ], "speakers": [], "host_url": "https://robotpad2026.github.io/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/RoboPAD", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/RoboPAD", "location": "Paris, France", "city": "Paris", "workshop_date": "2026-12-12", "contact": "linsihao6@gmail.com", "tracks": [ { "key": "RoboPAD", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/RoboPAD", "submission_dates_raw": "" } ], "group": "agents", "group_label": "Agents & Agentic Systems", "corpus": "RoboPAD: Post-Training Adaptation of Robot Foundation Models NeurIPS 2026 RoboPAD A NeurIPS 2026 workshop consolidating post-training - adaptation, correction, evaluation, and safe improvement of pretrained robot foundation models - as a shared research agenda, bringing together robot learning, reinforcement learning, foundation models, world models, and embodied AI. Human Feedback / Intervention (human corrections, teleoperation, preference learning) Policy Optimization / Specialization (RL fine-tuning, action decoders, fast specialization) World Model / Simulation (predictive models, simulation data, closed-loop refinement) Reasoning & Memory (deliberative reasoning, task decomposition, recovery) Cross-Embodiment Adaptation (humanoids, dexterous hands, sim-to-real transfer) Evaluation & Safety (benchmarks, reproducible protocols, constraint-respecting fine-tuning) About the Workshop\n\nPretrained robot foundation models are increasingly capable, yet real-world deployment still requires adaptation to new environments, embodiments, tasks, and failures. RoboPAD focuses on post-training: how to adapt, correct, evaluate, and safely improve these models after pretraining. Post-training is the algorithmic stage after pretraining: the closed-loop process of adapting, correcting, evaluating, and improving robot foundation models under real-world constraints.\n\nWe bring together researchers in robot learning, reinforcement learning, foundation models, world models, and embodied AI to consolidate post-training as a shared research agenda.\n\nTopics\n- Human Feedback / Intervention - Feedback- and intervention-driven adaptation: human corrections, teleoperation, preference learning.\n- Policy Optimization / Specialization - Policy optimization after pretraining: RL fine-tuning, action decoders, fast specialization.\n- World Model / Simulation - World-model and simulation-driven adaptation: predictive models, simulation data, closed-loop refinement.\n- Reasoning & Memory - Reasoning, memory, and agentic adaptation: deliberative reasoning, task decomposition, recovery.\n- Cross-Embodiment Adaptation - Cross-embodiment and whole-body adaptation: humanoids, dexterous hands, sim-to-real transfer.\n- Evaluation & Safety - Evaluation, safety, and robustness: benchmarks, reproducible protocols, constraint-respecting fine-tuning.\n\nCall for Papers\n\nWe welcome work in progress, under-review work, and unpublished or preprint work substantially reframed as open-problem, benchmark, or perspective contributions. Archival published papers are not eligible. Dual submissions must comply with other venues' policies.\n\nSubmission Tracks\n- Short papers (up to 4 pages) - preliminary findings, positions, benchmarks, infrastructure reports, and negative results.\n- Long papers (up to 9 pages) - mature contributions with fuller technical development, experiments, or analysis.\nPage limits exclude references and appendices.\n\nReview & status\nDouble-blind via OpenReview, with three reviewers per paper; non-archival (accepted papers posted on OpenReview unless authors opt out); posters + spotlights + Best Paper. At least one author must present in person.\n\nImportant Dates\n- Submission: Aug 29, 2026 (Anywhere on Earth / AoE)\n- Notification: Sep 29, 2026 (Anywhere on Earth / AoE)\n- Workshop: Dec 12-13, 2026, Paris, France\n\nAssociated Challenge - RoboWorld Challenge 2026\nRoboWorld 2026 is an independently organized challenge associated with the RoboPAD Workshop at NeurIPS 2026. It focuses on embodied world modeling for robotics and autonomous driving across multiple tracks. Challenge logistics, registration, and evaluation are managed by the RoboWorld organizing team; participation is separate from RoboPAD paper submission." }, { "key": "Robotics_World_Modeling", "title": "Robot Learning with World Models: Capabilities, Frontiers, and Challenges", "subtitle": "NeurIPS 2026 Robotics World Modeling Workshop", "summary": "This workshop explores using world models to advance Physical AI and address key challenges in robot learning, reasoning, and control, focusing on how predictive models of environment dynamics can bridge the gap between simulated dynamics and real-world physical interactions.", "cfp_full": "Robot Learning with World Models: Capabilities, Frontiers, and Challenges\n\nThis workshop explores using world models to advance Physical AI and address key challenges in robot learning, reasoning, and control. Recent breakthroughs in world modeling (e.g., Genie 3, Cosmos, Cosmos 3) have spurred significant progress in enabling robots to reason (e.g., Du et al. 2024, Physical Intelligence 2026), learn (e.g., Hafner et al. 2025), and be evaluated (e.g., Gemini Robotics Team 2025) in imagined scenarios. Furthermore, integrating spatiotemporal physical dynamics into control policies (World Action Models) has led to the emergence of zero-shot and visuomotor control policies (e.g., DreamZero, Cosmos Policy).\n\nHowever, state-of-the-art world models are primarily vision-focused and function essentially as controllable video generation models with plausible physical dynamics. The real world is far more complex, and fundamental challenges persist in consistency, physical accuracy, and the lack of multi-modalities for physical interactions.\n\nThe goal of this workshop is to bring together researchers and practitioners working at the frontiers of developing Physical AI with world models. We aim to exchange ideas on the state-of-the-art of controllable video models and discuss the critical challenges the community faces in bridging the gap between simulated dynamics and real-world physical interactions.\n\nCall For Papers\n\nWe invite the submission of research papers, position papers, and demo proposals on the topic of world models for robot learning. This workshop focuses on the intersection of world models and robotics, exploring how predictive models of environment dynamics can advance Physical AI.\n\nTopics of interest include, but are not limited to:\n- World Action Models (WAMs): Unifying world dynamics, spatiotemporal understanding, and robot action generation.\n- Learning, reasoning, and evaluation with imagined rollouts: Leveraging latent/observation space rollouts for planning, causal reasoning, and safety.\n- Multi-modality beyond vision: Integrating tactile sensing, proprioception, force feedback, audio, and other modalities.\n- Physical accuracy and spatiotemporal consistency: Simulating stable reality, realistic dynamics, and reliable control.\n- Evaluation metrics and benchmarks: Action-conditioned metrics and standardized benchmarks focusing on physical plausibility.\n\nSubmission Types:\n- Proposals for demo and networking: 1 page, with a light-weight and casual form, about anything that are related to robot learning with world models. We expect the proposal to have (1) What/How you would like to show or discuss - it can be simple as introducing what your lab is brewing or just a topic that you'd like to discuss with other attendees; (2) Optionally, what you would like to demo. The accepted proposals will have a space at the Demo and Networking session. Please submit 1 page PDF via the Google Form.\n- Full Papers: Up to 8 pages in NeurIPS or ICLR format, with potentially large-scale experiments (submit through OpenReview).\n- Short Papers: 2-4 pages in NeurIPS or ICLR format, with proof-of-concept demonstrations (submit through OpenReview).\n\nAccepted papers will be presented during poster sessions, with exceptional submissions selected for spotlight oral presentations. All accepted papers will be made publicly available as non-archival reports, allowing for future submissions to archival conferences or journals.\n\nImportant Dates:\n- Submission Deadline: August 29, 2026, AoE\n- Author Notification: September 25, 2026, AoE\n- Camera Ready Deadline: November 30, 2026, AoE\n- Workshop Date: December 11 or 12, 2026 (TBD)\n\nCamera Ready Instructions: Please incorporate reviewers' feedback and prepare your camera-ready submission on OpenReview. Your camera-ready submission should be de-anonymized and include at most 8 pages for full papers, and 2-4 pages for short papers, excluding references and appendices. The paper format must follow the NeurIPS style template. The camera-ready deadline is November 30, 2026, Anywhere on Earth (AoE).", "cfp_status": "published", "topics": [ "World Action Models (WAMs)", "Learning, reasoning, and evaluation with imagined rollouts", "Multi-modality beyond vision (tactile, proprioception, force feedback, audio)", "Physical accuracy and spatiotemporal consistency", "Evaluation metrics and benchmarks" ], "important_dates": [ { "label": "Submission Deadline (AoE)", "date": "2026-08-29" }, { "label": "Author Notification (AoE)", "date": "2026-09-25" }, { "label": "Camera Ready Deadline (AoE)", "date": "2026-11-30" }, { "label": "Workshop Date", "date": "December 11 or 12, 2026" } ], "organizers": [ "Homanga Bharadhwaj (Johns Hopkins University)", "Jing-Wen Chen (National Taiwan University)", "Hiroki Furuta (Google DeepMind / University of Tokyo)", "Kuang-Huei Lee (Google DeepMind)", "Ruoshi Liu (Amazon FAR)", "Zeyi Liu (Stanford University)", "Yifu Qiu (University of Edinburgh / University of Cambridge)", "Wenhao Yu (Google DeepMind)" ], "speakers": [ "Kristen Grauman (UT Austin)", "Jie Tan (Google DeepMind)", "Roberto Calandra (TU Dresden)", "Tsung-Yi Lin (NVIDIA Research)", "Hang Zhao (Tsinghua University / Galaxia AI)", "Wei-Chiu Ma (Cornell University)", "Franziska Meier (Waymo)" ], "host_url": "https://robowm-ws.github.io/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/Robotics_World_Modeling", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/Robotics_World_Modeling", "location": "Sydney", "city": "Sydney", "workshop_date": "", "contact": "pc@rwm.org", "tracks": [ { "key": "Robotics_World_Modeling", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/Robotics_World_Modeling", "submission_dates_raw": "Submission Deadline: Aug 22 2026 11:59AM UTC-0" } ], "group": "agents", "group_label": "Agents & Agentic Systems", "corpus": "Robot Learning with World Models: Capabilities, Frontiers, and Challenges NeurIPS 2026 Robotics World Modeling Workshop This workshop explores using world models to advance Physical AI and address key challenges in robot learning, reasoning, and control, focusing on how predictive models of environment dynamics can bridge the gap between simulated dynamics and real-world physical interactions. World Action Models (WAMs) Learning, reasoning, and evaluation with imagined rollouts Multi-modality beyond vision (tactile, proprioception, force feedback, audio) Physical accuracy and spatiotemporal consistency Evaluation metrics and benchmarks Robot Learning with World Models: Capabilities, Frontiers, and Challenges\n\nThis workshop explores using world models to advance Physical AI and address key challenges in robot learning, reasoning, and control. Recent breakthroughs in world modeling (e.g., Genie 3, Cosmos, Cosmos 3) have spurred significant progress in enabling robots to reason (e.g., Du et al. 2024, Physical Intelligence 2026), learn (e.g., Hafner et al. 2025), and be evaluated (e.g., Gemini Robotics Team 2025) in imagined scenarios. Furthermore, integrating spatiotemporal physical dynamics into control policies (World Action Models) has led to the emergence of zero-shot and visuomotor control policies (e.g., DreamZero, Cosmos Policy).\n\nHowever, state-of-the-art world models are primarily vision-focused and function essentially as controllable video generation models with plausible physical dynamics. The real world is far more complex, and fundamental challenges persist in consistency, physical accuracy, and the lack of multi-modalities for physical interactions.\n\nThe goal of this workshop is to bring together researchers and practitioners working at the frontiers of developing Physical AI with world models. We aim to exchange ideas on the state-of-the-art of controllable video models and discuss the critical challenges the community faces in bridging the gap between simulated dynamics and real-world physical interactions.\n\nCall For Papers\n\nWe invite the submission of research papers, position papers, and demo proposals on the topic of world models for robot learning. This workshop focuses on the intersection of world models and robotics, exploring how predictive models of environment dynamics can advance Physical AI.\n\nTopics of interest include, but are not limited to:\n- World Action Models (WAMs): Unifying world dynamics, spatiotemporal understanding, and robot action generation.\n- Learning, reasoning, and evaluation with imagined rollouts: Leveraging latent/observation space rollouts for planning, causal reasoning, and safety.\n- Multi-modality beyond vision: Integrating tactile sensing, proprioception, force feedback, audio, and other modalities.\n- Physical accuracy and spatiotemporal consistency: Simulating stable reality, realistic dynamics, and reliable control.\n- Evaluation metrics and benchmarks: Action-conditioned metrics and standardized benchmarks focusing on physical plausibility.\n\nSubmission Types:\n- Proposals for demo and networking: 1 page, with a light-weight and casual form, about anything that are related to robot learning with world models. We expect the proposal to have (1) What/How you would like to show or discuss - it can be simple as introducing what your lab is brewing or just a topic that you'd like to discuss with other attendees; (2) Optionally, what you would like to demo. The accepted proposals will have a space at the Demo and Networking session. Please submit 1 page PDF via the Google Form.\n- Full Papers: Up to 8 pages in NeurIPS or ICLR format, with potentially large-scale experiments (submit through OpenReview).\n- Short Papers: 2-4 pages in NeurIPS or ICLR format, with proof-of-concept demonstrations (submit through OpenReview).\n\nAccepted papers will be presented during poster sessions, with exceptional submissions selected for spotlight oral presentations. All accepted papers will be made publicly available as non-archival reports, allowing for future submissions to archival conferences or journals.\n\nImportant Dates:\n- Submission Deadline: August 29, 2026, AoE\n- Author Notification: September 25, 2026, AoE\n- Camera Ready Deadline: November 30, 2026, AoE\n- Workshop Date: December 11 or 12, 2026 (TBD)\n\nCamera Ready Instructions: Please incorporate reviewers' feedback and prepare your camera-ready submission on OpenReview. Your camera-ready submission should be de-anonymized and include at most 8 pages for full papers, and 2-4 pages for short papers, excluding references and appendices. The paper format must follow the NeurIPS style template. The camera-ready deadline is November 30, 2026, Anywhere on Earth (AoE)." }, { "key": "RoCo-Spring", "title": "RoCo-Spring: The Robust Correspondence Challenge", "subtitle": "RoCo-Spring 2026", "summary": "RoCo-Spring is a NeurIPS 2026 Competition Track challenge on robust dense correspondence under realistic distribution shifts, jointly evaluating clean accuracy and corrupted robustness across optical flow, stereo matching, and scene flow, with an accompanying exploration track.", "cfp_full": "RoCo-Spring: The Robust Correspondence Challenge\n\nOverview\nA NeurIPS 2026 Challenge on robust dense correspondence under realistic distribution shifts, covering optical flow, stereo matching, and scene flow.\n\nDense correspondence has made remarkable progress on standard benchmarks. Although modern optical flow, stereo matching, and scene flow methods achieve strong performance under ideal conditions, their accuracy can degrade substantially under real-world visual shifts.\n\nRobustness under realistic corruptions remains a challenge for the deployment of these methods in autonomous driving, robotics, and other safety-critical settings. RoCo-Spring jointly evaluates clean accuracy and corrupted robustness, targeting methods that remain accurate under camera noise, adverse weather conditions, blur, compression, and changes in illumination.\n\nDatasets: Spring and RobustSpring\n\nSpring\nSpring provides high-resolution stereo video with left and right images at 1920x1080 px, together with dense scene-flow ground truth at 4x super-resolution. It includes bidirectional disparities for stereo and depth, disparity changes for scene flow, and forward/backward optical flow for both left and right views. This makes Spring a unified high-detail benchmark for optical flow, stereo matching, and scene flow.\n\nRobustSpring\nRobustSpring extends Spring into a robustness benchmark with 20 realistic image corruptions, including blur, color changes, noise, quality degradations, and weather effects. The corruptions are applied to stereo video data and are integrated consistently over time, across stereo views, and with depth where applicable, enabling controlled robustness evaluation for optical flow, stereo, and scene flow.\n\nTracks: Four challenge tracks\n\nOptical Flow\nEstimate a dense 2D displacement field between consecutive frames, assigning each visible pixel a horizontal and vertical motion vector.\n\nStereo Matching\nEstimate a dense disparity map from a rectified stereo image pair, assigning each pixel the horizontal offset to its corresponding point.\n\nScene Flow\nEstimate dense 3D motion from stereo image sequences, combining per-pixel geometry and temporal motion into a single correspondence task.\n\nExploration Track\nSubmit a rigorous analysis of robustness in dense correspondence, focusing on failure modes, method behavior, metrics, or evaluation design.\n\nTimeline: Competition schedule\n- July 2, 2026 - Website launch\n- July 13, 2026 - Development leaderboard opens for all quantitative tasks\n- September 15, 2026 - 4-6 page workshop paper submission deadline\n- September 29, 2026 - Workshop paper author notification\n- September 30, 2026 - Final quantitative submission deadline\n- October 7, 2026 - Camera-ready paper, code, and reproducibility package deadline\n- October 15, 2026 - Final evaluation, reproducibility checks, and award shortlisting\n- October 31, 2026 - Winners notified and workshop program finalized\n- December 11 or 12, 2026 - In-person NeurIPS Competition Track workshop\nAll dates are tentative and subject to change.\n\nParticipation: OpenReview submission, evaluation, and a starter kit (codebase) are available. Registration is required to register or manage your team. Leaderboards coming soon. Starter kit codebase: hmorimitsu/roco-spring-devkit.\n\nContact\nChallenge email: roco-spring-org@googlegroups.com", "cfp_status": "published", "topics": [ "Optical flow (dense 2D displacement field between consecutive frames)", "Stereo matching (dense disparity map from rectified stereo image pairs)", "Scene flow (dense 3D motion from stereo image sequences)", "Robustness of dense correspondence under realistic corruptions", "Camera noise, adverse weather, blur, compression, and illumination changes", "Exploration/analysis of failure modes, method behavior, metrics, and evaluation design" ], "important_dates": [ { "label": "Website launch", "date": "2026-07-02" }, { "label": "Development leaderboard opens", "date": "2026-07-13" }, { "label": "Workshop paper submission deadline (4-6 pages)", "date": "2026-09-15" }, { "label": "Workshop paper author notification", "date": "2026-09-29" }, { "label": "Final quantitative submission deadline", "date": "2026-09-30" }, { "label": "Camera-ready paper, code, and reproducibility package deadline", "date": "2026-10-07" }, { "label": "Final evaluation, reproducibility checks, and award shortlisting", "date": "2026-10-15" }, { "label": "Winners notified and workshop program finalized", "date": "2026-10-31" }, { "label": "In-person NeurIPS Competition Track workshop", "date": "2026-12-11" } ], "organizers": [ "Shashank Agnihotri (University of Mannheim)", "Victor Oei (University of Stuttgart)", "Jenny Schmalfuss (NVIDIA)", "Katrin Bauer (University of Stuttgart)", "Henrique Morimitsu (University of Science and Technology Beijing)", "Andrés Bruhn (University of Stuttgart)", "Margret Keuper (University of Mannheim and MPI-INF)" ], "speakers": [ "Jia Deng (Princeton University, Princeton Vision & Learning Lab)", "Fatih Porikli (Qualcomm AI Research)" ], "host_url": "https://roco-spring.github.io/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/RoCo-Spring", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/RoCo-Spring", "location": "Sydney, NuerIPS 2026", "city": "Sydney", "workshop_date": "2026-12-10", "contact": "shashank.agnihotri@uni-mannheim.de", "tracks": [ { "key": "RoCo-Spring", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/RoCo-Spring", "submission_dates_raw": "" } ], "group": "theory", "group_label": "Theory & Methods", "corpus": "RoCo-Spring: The Robust Correspondence Challenge RoCo-Spring 2026 RoCo-Spring is a NeurIPS 2026 Competition Track challenge on robust dense correspondence under realistic distribution shifts, jointly evaluating clean accuracy and corrupted robustness across optical flow, stereo matching, and scene flow, with an accompanying exploration track. Optical flow (dense 2D displacement field between consecutive frames) Stereo matching (dense disparity map from rectified stereo image pairs) Scene flow (dense 3D motion from stereo image sequences) Robustness of dense correspondence under realistic corruptions Camera noise, adverse weather, blur, compression, and illumination changes Exploration/analysis of failure modes, method behavior, metrics, and evaluation design RoCo-Spring: The Robust Correspondence Challenge\n\nOverview\nA NeurIPS 2026 Challenge on robust dense correspondence under realistic distribution shifts, covering optical flow, stereo matching, and scene flow.\n\nDense correspondence has made remarkable progress on standard benchmarks. Although modern optical flow, stereo matching, and scene flow methods achieve strong performance under ideal conditions, their accuracy can degrade substantially under real-world visual shifts.\n\nRobustness under realistic corruptions remains a challenge for the deployment of these methods in autonomous driving, robotics, and other safety-critical settings. RoCo-Spring jointly evaluates clean accuracy and corrupted robustness, targeting methods that remain accurate under camera noise, adverse weather conditions, blur, compression, and changes in illumination.\n\nDatasets: Spring and RobustSpring\n\nSpring\nSpring provides high-resolution stereo video with left and right images at 1920x1080 px, together with dense scene-flow ground truth at 4x super-resolution. It includes bidirectional disparities for stereo and depth, disparity changes for scene flow, and forward/backward optical flow for both left and right views. This makes Spring a unified high-detail benchmark for optical flow, stereo matching, and scene flow.\n\nRobustSpring\nRobustSpring extends Spring into a robustness benchmark with 20 realistic image corruptions, including blur, color changes, noise, quality degradations, and weather effects. The corruptions are applied to stereo video data and are integrated consistently over time, across stereo views, and with depth where applicable, enabling controlled robustness evaluation for optical flow, stereo, and scene flow.\n\nTracks: Four challenge tracks\n\nOptical Flow\nEstimate a dense 2D displacement field between consecutive frames, assigning each visible pixel a horizontal and vertical motion vector.\n\nStereo Matching\nEstimate a dense disparity map from a rectified stereo image pair, assigning each pixel the horizontal offset to its corresponding point.\n\nScene Flow\nEstimate dense 3D motion from stereo image sequences, combining per-pixel geometry and temporal motion into a single correspondence task.\n\nExploration Track\nSubmit a rigorous analysis of robustness in dense correspondence, focusing on failure modes, method behavior, metrics, or evaluation design.\n\nTimeline: Competition schedule\n- July 2, 2026 - Website launch\n- July 13, 2026 - Development leaderboard opens for all quantitative tasks\n- September 15, 2026 - 4-6 page workshop paper submission deadline\n- September 29, 2026 - Workshop paper author notification\n- September 30, 2026 - Final quantitative submission deadline\n- October 7, 2026 - Camera-ready paper, code, and reproducibility package deadline\n- October 15, 2026 - Final evaluation, reproducibility checks, and award shortlisting\n- October 31, 2026 - Winners notified and workshop program finalized\n- December 11 or 12, 2026 - In-person NeurIPS Competition Track workshop\nAll dates are tentative and subject to change.\n\nParticipation: OpenReview submission, evaluation, and a starter kit (codebase) are available. Registration is required to register or manage your team. Leaderboards coming soon. Starter kit codebase: hmorimitsu/roco-spring-devkit.\n\nContact\nChallenge email: roco-spring-org@googlegroups.com" }, { "key": "SocialAgent", "title": "Second Workshop on Large Language Models for Social Reasoning and Simulation", "subtitle": "SocialAgent@NeurIPS 2026", "summary": "SocialAgent brings together researchers who build, evaluate, and critically examine LLM-based social agents that reason, interact, and generate behavior in simulated environments, advancing a principled, responsible, and empirically grounded research agenda for social reasoning and simulation.", "cfp_full": "SocialAgent: Second Workshop on Large Language Models for Social Reasoning and Simulation\nNeurIPS 2026 Workshop\nAtlanta, Georgia, USA - December 12 or 13, 2026 (Exact Date TBA)\n\nOverview\nLarge language models are increasingly used not only as analytic tools, but as socially situated agents that reason, interact, and generate behavior in simulated environments. This shift enables new forms of research on decision-making, social norms, cooperation, persuasion, and collective dynamics at scale.\n\nSocialAgent brings together researchers who build, evaluate, and critically examine LLM-based social agents. The workshop aims to advance a principled, responsible, and empirically grounded research agenda for social reasoning and simulation.\n\nWorkshop Themes\n- Social reasoning & cognition: Theory of mind, moral judgment, and genuine reasoning vs. prompting artifacts.\n- Social simulation & its validity: LLMs as proxies for individuals and populations, and the validity standards that make simulations trustworthy.\n- Pluralistic alignment & evaluation: Contested values, emergent norms, and evaluation beyond task accuracy.\n\nCall for Papers\nWe welcome empirical, methodological, theoretical, and conceptual submissions. Topics include, but are not limited to:\n- Agent-Based Social Simulation with LLMs: Coordination, conflict, cooperation, collective action, and information diffusion.\n- Emergence, Norm Formation, and Cultural Evolution: The development of norms, beliefs, narratives, and collective behavior.\n- Persona Fidelity and Population Heterogeneity: Psychologically plausible agents, synthetic populations, and demographic variation.\n- Platform and Network Contexts: Exposure structures, feedback loops, incentives, moderation, and governance.\n- Interventions and Counterfactual Experiments: Simulations for policy, system design, and stress testing.\n- Grounding, Realism, and Evaluation: Calibration, reproducibility, privacy, and comparisons with human behavior.\n- Ethical and Societal Implications: Validity limits, misuse risks, and responsible applications in sensitive domains.\n\nSubmission Types and Guidelines\n- Long Paper (Non-Archival): May consist of up to 9 pages of content, plus unlimited pages for references and appendix.\n- Short / Position / Abstract Papers (Non-Archival): May consist of up to 4 pages of content, plus unlimited references and appendix.\n\nSubmission Policies\n- The reviewing process will be double-blind.\n- At least one author of each accepted paper should register and present their work in person at SocialAgent.\n\nCall for Reviewers\nWe invite researchers working on large language models, social reasoning, agent-based simulation, and related areas to join our reviewer pool.\n\nImportant Dates\n- Submission Deadline: August 25, 2026, AoE\n- Author Notification: September 29, 2026, AoE\n- Workshop Date: December 12 and December 13, 2026, Atlanta TBD\n\nContact\nFor questions, contact us through the SocialAgent Slack channel or email social.llm.workshop@gmail.com.", "cfp_status": "published", "topics": [ "Agent-Based Social Simulation with LLMs: coordination, conflict, cooperation, collective action, and information diffusion", "Emergence, Norm Formation, and Cultural Evolution: development of norms, beliefs, narratives, and collective behavior", "Persona Fidelity and Population Heterogeneity: psychologically plausible agents, synthetic populations, and demographic variation", "Platform and Network Contexts: exposure structures, feedback loops, incentives, moderation, and governance", "Interventions and Counterfactual Experiments: simulations for policy, system design, and stress testing", "Grounding, Realism, and Evaluation: calibration, reproducibility, privacy, and comparisons with human behavior", "Ethical and Societal Implications: validity limits, misuse risks, and responsible applications in sensitive domains", "Social reasoning & cognition: theory of mind, moral judgment, and reasoning vs. prompting artifacts", "Pluralistic alignment & evaluation: contested values, emergent norms, and evaluation beyond task accuracy" ], "important_dates": [ { "label": "Submission Deadline (AoE)", "date": "2026-08-25" }, { "label": "Author Notification (AoE)", "date": "2026-09-29" }, { "label": "Workshop", "date": "2026-12-12" } ], "organizers": [ "Xiangjue Dong (TAMU & Microsoft)", "Jiseon Kim (Vector Institute)", "Yunah Jang (Seoul National University)", "Tim G. J. Rudner (University of Toronto & Vector Institute)", "Alice Oh (KAIST)", "Siqi Zhang (Microsoft, Technical Support)" ], "speakers": [ "Serina Chang (UC Berkeley)", "Afra Mashhadi (University of Washington)" ], "host_url": "https://social-llm-workshop.github.io/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/SocialAgent", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/SocialAgent", "location": "Atlanta, USA", "city": "Atlanta", "workshop_date": "2026-12-12", "contact": "social.llm.workshop@gmail.com", "tracks": [ { "key": "SocialAgent", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/SocialAgent", "submission_dates_raw": "" } ], "group": "society", "group_label": "Society, Applications & Community", "corpus": "Second Workshop on Large Language Models for Social Reasoning and Simulation SocialAgent@NeurIPS 2026 SocialAgent brings together researchers who build, evaluate, and critically examine LLM-based social agents that reason, interact, and generate behavior in simulated environments, advancing a principled, responsible, and empirically grounded research agenda for social reasoning and simulation. Agent-Based Social Simulation with LLMs: coordination, conflict, cooperation, collective action, and information diffusion Emergence, Norm Formation, and Cultural Evolution: development of norms, beliefs, narratives, and collective behavior Persona Fidelity and Population Heterogeneity: psychologically plausible agents, synthetic populations, and demographic variation Platform and Network Contexts: exposure structures, feedback loops, incentives, moderation, and governance Interventions and Counterfactual Experiments: simulations for policy, system design, and stress testing Grounding, Realism, and Evaluation: calibration, reproducibility, privacy, and comparisons with human behavior Ethical and Societal Implications: validity limits, misuse risks, and responsible applications in sensitive domains Social reasoning & cognition: theory of mind, moral judgment, and reasoning vs. prompting artifacts Pluralistic alignment & evaluation: contested values, emergent norms, and evaluation beyond task accuracy SocialAgent: Second Workshop on Large Language Models for Social Reasoning and Simulation\nNeurIPS 2026 Workshop\nAtlanta, Georgia, USA - December 12 or 13, 2026 (Exact Date TBA)\n\nOverview\nLarge language models are increasingly used not only as analytic tools, but as socially situated agents that reason, interact, and generate behavior in simulated environments. This shift enables new forms of research on decision-making, social norms, cooperation, persuasion, and collective dynamics at scale.\n\nSocialAgent brings together researchers who build, evaluate, and critically examine LLM-based social agents. The workshop aims to advance a principled, responsible, and empirically grounded research agenda for social reasoning and simulation.\n\nWorkshop Themes\n- Social reasoning & cognition: Theory of mind, moral judgment, and genuine reasoning vs. prompting artifacts.\n- Social simulation & its validity: LLMs as proxies for individuals and populations, and the validity standards that make simulations trustworthy.\n- Pluralistic alignment & evaluation: Contested values, emergent norms, and evaluation beyond task accuracy.\n\nCall for Papers\nWe welcome empirical, methodological, theoretical, and conceptual submissions. Topics include, but are not limited to:\n- Agent-Based Social Simulation with LLMs: Coordination, conflict, cooperation, collective action, and information diffusion.\n- Emergence, Norm Formation, and Cultural Evolution: The development of norms, beliefs, narratives, and collective behavior.\n- Persona Fidelity and Population Heterogeneity: Psychologically plausible agents, synthetic populations, and demographic variation.\n- Platform and Network Contexts: Exposure structures, feedback loops, incentives, moderation, and governance.\n- Interventions and Counterfactual Experiments: Simulations for policy, system design, and stress testing.\n- Grounding, Realism, and Evaluation: Calibration, reproducibility, privacy, and comparisons with human behavior.\n- Ethical and Societal Implications: Validity limits, misuse risks, and responsible applications in sensitive domains.\n\nSubmission Types and Guidelines\n- Long Paper (Non-Archival): May consist of up to 9 pages of content, plus unlimited pages for references and appendix.\n- Short / Position / Abstract Papers (Non-Archival): May consist of up to 4 pages of content, plus unlimited references and appendix.\n\nSubmission Policies\n- The reviewing process will be double-blind.\n- At least one author of each accepted paper should register and present their work in person at SocialAgent.\n\nCall for Reviewers\nWe invite researchers working on large language models, social reasoning, agent-based simulation, and related areas to join our reviewer pool.\n\nImportant Dates\n- Submission Deadline: August 25, 2026, AoE\n- Author Notification: September 29, 2026, AoE\n- Workshop Date: December 12 and December 13, 2026, Atlanta TBD\n\nContact\nFor questions, contact us through the SocialAgent Slack channel or email social.llm.workshop@gmail.com." }, { "key": "Sim2Sci", "title": "Sim2Science: ML with Imperfect Scientific Models", "subtitle": "Sim2Sci 2026", "summary": "Sim2Science is a cross-domain workshop on machine learning for imperfect, misspecified scientific simulators, asking how ML can best leverage and mitigate the limitations of mechanistic models across chemistry, fusion, neuroscience, climate, and beyond.", "cfp_full": "About the Workshop\n\nAI4Science has matured into an established field, with ML now embedded throughout the simulator-based workflows of the natural sciences. Much of this progress runs through simulators—mechanistic models hand-crafted by domain experts and fit to data—that encode our scientific theories and underpin prediction, parameter inference, experimentation, and decision-making. Yet an ML method coupled to a simulator is only as good as that simulator: simulators simplify complex systems, omit intractable physics, and depend on uncertain parameters, creating a discrepancy between simulated and observed data that biases the scientific conclusions we draw.\n\nThe central question of this workshop is: How can we best leverage imperfect scientific simulators when confronted with real-world data, and how can ML help to account for and mitigate limitations in simulator-based workflows across a wide range of domains? Sim2Science is deliberately cross-domain: rather than focusing on a single scientific field, we bring together researchers who each maintain hierarchies of simulators at different fidelities—in chemistry, fusion, neuroscience, climate, and beyond—to build a shared vocabulary and toolkit for handling imperfect simulators, so that progress in one field can transfer to others.\n\nTopics of Interest\n\nWe welcome contributions of any kind — new methods, applications, analyses, benchmarks, or position pieces — spanning biology, chemistry, physics, materials science, climate science, and related fields, as long as the work engages both machine learning and scientific simulators. Topics include:\n- Simulation-based inference and related parameter inference methods\n- Understanding and mitigating model misspecification, including simulator diagnostics and discrepancy modeling\n- Emulator and surrogate modeling, as well as hybrid and physics-informed approaches\n- Analysis of simulator structure, degeneracy, simplifications, and identifiability\n- Simulator pipelines, including data handling, preprocessing, and integration with downstream ML models\n- Active learning and Bayesian optimization for fitting parameters or model components\n- Closed-loop and experiment-in-the-loop scientific workflows\n- Multi-fidelity and multi-resolution modeling\n- (Agentic) model and equation discovery\n- Differentiable frameworks, LLM-assisted scientific reasoning, and workflow automation\n\nFull submission tracks, instructions, and example simulators are on the Call for Papers page.\n\nImportant Dates (All deadlines are 23:59 Anywhere on Earth (AoE)):\n- Submission Deadline: August 29, 2026\n- Author Notification: September 29, 2026\n- Camera-Ready Deadline: TBD (shortly before the workshop)\n- Workshop Date: December 12 or 13, 2026, Paris, France", "cfp_status": "published", "topics": [ "Simulation-based inference and related parameter inference methods", "Understanding and mitigating model misspecification, including simulator diagnostics and discrepancy modeling", "Emulator and surrogate modeling, as well as hybrid and physics-informed approaches", "Analysis of simulator structure, degeneracy, simplifications, and identifiability", "Simulator pipelines, including data handling, preprocessing, and integration with downstream ML models", "Active learning and Bayesian optimization for fitting parameters or model components", "Closed-loop and experiment-in-the-loop scientific workflows", "Multi-fidelity and multi-resolution modeling", "(Agentic) model and equation discovery", "Differentiable frameworks, LLM-assisted scientific reasoning, and workflow automation" ], "important_dates": [ { "label": "Submission Deadline", "date": "2026-08-29" }, { "label": "Author Notification", "date": "2026-09-29" }, { "label": "Camera-Ready Deadline", "date": "TBD (shortly before the workshop)" }, { "label": "Workshop Date", "date": "December 12 or 13, 2026" } ], "organizers": [], "speakers": [], "host_url": "https://www.sim2science.com/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/Sim2Sci", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/Sim2Sci", "location": "Paris, France", "city": "Paris", "workshop_date": "2026-12-10", "contact": "sim2science@gmail.com", "tracks": [ { "key": "Sim2Sci", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/Sim2Sci", "submission_dates_raw": "" } ], "group": "science", "group_label": "AI for Science", "corpus": "Sim2Science: ML with Imperfect Scientific Models Sim2Sci 2026 Sim2Science is a cross-domain workshop on machine learning for imperfect, misspecified scientific simulators, asking how ML can best leverage and mitigate the limitations of mechanistic models across chemistry, fusion, neuroscience, climate, and beyond. Simulation-based inference and related parameter inference methods Understanding and mitigating model misspecification, including simulator diagnostics and discrepancy modeling Emulator and surrogate modeling, as well as hybrid and physics-informed approaches Analysis of simulator structure, degeneracy, simplifications, and identifiability Simulator pipelines, including data handling, preprocessing, and integration with downstream ML models Active learning and Bayesian optimization for fitting parameters or model components Closed-loop and experiment-in-the-loop scientific workflows Multi-fidelity and multi-resolution modeling (Agentic) model and equation discovery Differentiable frameworks, LLM-assisted scientific reasoning, and workflow automation About the Workshop\n\nAI4Science has matured into an established field, with ML now embedded throughout the simulator-based workflows of the natural sciences. Much of this progress runs through simulators—mechanistic models hand-crafted by domain experts and fit to data—that encode our scientific theories and underpin prediction, parameter inference, experimentation, and decision-making. Yet an ML method coupled to a simulator is only as good as that simulator: simulators simplify complex systems, omit intractable physics, and depend on uncertain parameters, creating a discrepancy between simulated and observed data that biases the scientific conclusions we draw.\n\nThe central question of this workshop is: How can we best leverage imperfect scientific simulators when confronted with real-world data, and how can ML help to account for and mitigate limitations in simulator-based workflows across a wide range of domains? Sim2Science is deliberately cross-domain: rather than focusing on a single scientific field, we bring together researchers who each maintain hierarchies of simulators at different fidelities—in chemistry, fusion, neuroscience, climate, and beyond—to build a shared vocabulary and toolkit for handling imperfect simulators, so that progress in one field can transfer to others.\n\nTopics of Interest\n\nWe welcome contributions of any kind — new methods, applications, analyses, benchmarks, or position pieces — spanning biology, chemistry, physics, materials science, climate science, and related fields, as long as the work engages both machine learning and scientific simulators. Topics include:\n- Simulation-based inference and related parameter inference methods\n- Understanding and mitigating model misspecification, including simulator diagnostics and discrepancy modeling\n- Emulator and surrogate modeling, as well as hybrid and physics-informed approaches\n- Analysis of simulator structure, degeneracy, simplifications, and identifiability\n- Simulator pipelines, including data handling, preprocessing, and integration with downstream ML models\n- Active learning and Bayesian optimization for fitting parameters or model components\n- Closed-loop and experiment-in-the-loop scientific workflows\n- Multi-fidelity and multi-resolution modeling\n- (Agentic) model and equation discovery\n- Differentiable frameworks, LLM-assisted scientific reasoning, and workflow automation\n\nFull submission tracks, instructions, and example simulators are on the Call for Papers page.\n\nImportant Dates (All deadlines are 23:59 Anywhere on Earth (AoE)):\n- Submission Deadline: August 29, 2026\n- Author Notification: September 29, 2026\n- Camera-Ready Deadline: TBD (shortly before the workshop)\n- Workshop Date: December 12 or 13, 2026, Paris, France" }, { "key": "SLM-Agents", "title": "SLM-Agents: 1st NeurIPS Workshop on SLMs for Agentic Systems", "subtitle": "SLM-Agents", "summary": "This workshop is dedicated to small language models (SLMs) as the foundation of agentic AI systems, sitting at the intersection of efficient language model architectures and compression, agentic AI systems capable of autonomous reasoning and tool use, and edge computing and on-device deployment.", "cfp_full": "SLM-Agents: 1st Workshop on SLMs for Agentic Systems\nAccepted at NeurIPS 2026\nParis, France · December 12-13, 2026 (Exact workshop day to be announced)\n\nAbout the workshop\nScope and objectives\nThis workshop is dedicated to small language models (SLMs) as the foundation of agentic AI systems. Although large language models (LLMs) have demonstrated remarkable capabilities, their dependence on cloud infrastructure creates fundamental barriers to deployment in agentic pipelines—latency, privacy, connectivity, and substantial computational cost. SLMs offer a compelling alternative: recent studies argue that SLMs, not LLMs, might be a right option for the repetitive, narrowly scoped sub-tasks that dominate real agentic workloads. SLMs make it possible for autonomous AI agents to plan, reason, and act directly on resource-constrained devices such as smartphones, IoT systems, robotics platforms, and embedded systems. The workshop sits at the intersection of three rapidly evolving fields: (1) efficient language model architectures and compression techniques, (2) agentic AI systems capable of autonomous reasoning and tool use, and (3) edge computing and on-device deployment.\n\nOpen problems\nCompression and distillation: quantization, pruning, knowledge distillation, and architectural innovations for parameter-efficient LMs.\nHardware co-design: on-device inference optimization, NPU/accelerator-aware design, memory-bandwidth-bound serving, and energy-budgeted decoding.\nTraining for cooperation: fine-tuning SLMs for tool use, planning, multi-step reasoning, and small-large model handoff in heterogeneous agent stacks.\nEvaluation and benchmarks: task-success-per-watt, latency- and memory-aware leaderboards, and reproducible on-device evaluation harnesses.\nApplications and safety: privacy-preserving local processing, federated learning, and deployment case studies across mobile assistants, robotics, healthcare, automotive, and financial services, with associated safety, robustness, and provenance considerations.\n\nCall for Papers\nWe invite original work in progress on small language models for agentic systems.\n\nSubmission format\n- Extended abstracts: 4 pages plus references\n- Optional full papers: 8 pages plus references\n- NeurIPS workshop template\n- Double-blind review through OpenReview\n- Three reviewers per submission\n\nScope and publication\n- Original work in progress\n- Not under review at or accepted to the NeurIPS 2026 main program\n- Not previously published at a major ML or AI venue\n- Accepted papers hosted on OpenReview with author opt-in\n- Non-archival and not included in the NeurIPS proceedings\n\nTopics of interest\n- SLM architectures, training and inference\n- Agentic reasoning, planning and tool use\n- Hardware-aware and on-device deployment\n- Evaluation, benchmarks and efficiency\n- Privacy, safety, robustness and applications\n\nImportant Dates\nCall for Papers released: July 18, 2026\nSubmission portal opens: July 25, 2026\nPaper submission deadline: August 22, 2026 (AoE)\nReviewing period: August 30-September 19, 2026\nAcceptance notification: September 22, 2026\nCamera-ready deadline: October 13, 2026\nWorkshop: December 12 or 13, 2026\n\nEdge Agent Efficiency Challenge\nThe challenge focuses on running a multi-step agent task on consumer-class hardware, such as a laptop GPU, mobile NPU, or single-board computer, under fixed memory, latency, and energy budgets. Entries will be evaluated using a cost-adjusted task-success metric and will release reusable model checkpoints and inference recipes. Challenge winners will present their systems during the workshop's live-demo session.\n\nPolicies\nNon-archival workshop: Accepted workshop submissions are non-archival and may be submitted elsewhere after the workshop, subject to the policies of the other venue.\nParticipation and inclusion: The program includes mentorship for junior researchers, a dedicated lightning-talk track, outreach through affinity communities, and an inclusive workshop environment for participants across backgrounds and career stages.\n\nFor questions, contact the organizers at neurips.slmagents.2026@gmail.com.", "cfp_status": "published", "topics": [ "SLM architectures, training and inference", "Agentic reasoning, planning and tool use", "Hardware-aware and on-device deployment", "Evaluation, benchmarks and efficiency", "Privacy, safety, robustness and applications", "Compression and distillation (quantization, pruning, knowledge distillation)", "Hardware co-design and NPU/accelerator-aware design", "Training SLMs for cooperation and small-large model handoff" ], "important_dates": [ { "label": "Call for Papers released", "date": "2026-07-18" }, { "label": "Submission portal opens", "date": "2026-07-25" }, { "label": "Paper submission deadline", "date": "2026-08-22" }, { "label": "Reviewing period", "date": "August 30-September 19, 2026" }, { "label": "Acceptance notification", "date": "2026-09-22" }, { "label": "Camera-ready deadline", "date": "2026-10-13" }, { "label": "Workshop", "date": "December 12 or 13, 2026" } ], "organizers": [ "Habib Hajimolahoseini (AMD)", "Mehdi Rezagholizadeh (AMD)", "Vahid Partovi Nia (École Polytechnique de Montréal and Huawei Canada Noah's Ark Lab)", "Shahrzad Kianidehkordi (RBC Borealis)", "Mouloud Belbahri (TD Insurance)", "Pavlo Molchanov (NVIDIA Research)", "MirHamed Jafarzadeh Asl (Huawei Noah's Ark Lab)", "Walid Ahmed (Workday AI Research)" ], "speakers": [ "Emmanuel Abbe (EPFL and Apple)", "Ali Ghodsi (University of Waterloo)", "Diana Marculescu (The University of Texas at Austin)", "Amir Gholami (University of California, Berkeley)", "Layla El Asri (RBC Borealis)" ], "host_url": "https://slmw2026.github.io/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/SLM-Agents", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/SLM-Agents", "location": "Paris", "city": "Paris", "workshop_date": "2026-12-12", "contact": "neurips.slmagents.2026@gmail.com", "tracks": [ { "key": "SLM-Agents", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/SLM-Agents", "submission_dates_raw": "" } ], "group": "agents", "group_label": "Agents & Agentic Systems", "corpus": "SLM-Agents: 1st NeurIPS Workshop on SLMs for Agentic Systems SLM-Agents This workshop is dedicated to small language models (SLMs) as the foundation of agentic AI systems, sitting at the intersection of efficient language model architectures and compression, agentic AI systems capable of autonomous reasoning and tool use, and edge computing and on-device deployment. SLM architectures, training and inference Agentic reasoning, planning and tool use Hardware-aware and on-device deployment Evaluation, benchmarks and efficiency Privacy, safety, robustness and applications Compression and distillation (quantization, pruning, knowledge distillation) Hardware co-design and NPU/accelerator-aware design Training SLMs for cooperation and small-large model handoff SLM-Agents: 1st Workshop on SLMs for Agentic Systems\nAccepted at NeurIPS 2026\nParis, France · December 12-13, 2026 (Exact workshop day to be announced)\n\nAbout the workshop\nScope and objectives\nThis workshop is dedicated to small language models (SLMs) as the foundation of agentic AI systems. Although large language models (LLMs) have demonstrated remarkable capabilities, their dependence on cloud infrastructure creates fundamental barriers to deployment in agentic pipelines—latency, privacy, connectivity, and substantial computational cost. SLMs offer a compelling alternative: recent studies argue that SLMs, not LLMs, might be a right option for the repetitive, narrowly scoped sub-tasks that dominate real agentic workloads. SLMs make it possible for autonomous AI agents to plan, reason, and act directly on resource-constrained devices such as smartphones, IoT systems, robotics platforms, and embedded systems. The workshop sits at the intersection of three rapidly evolving fields: (1) efficient language model architectures and compression techniques, (2) agentic AI systems capable of autonomous reasoning and tool use, and (3) edge computing and on-device deployment.\n\nOpen problems\nCompression and distillation: quantization, pruning, knowledge distillation, and architectural innovations for parameter-efficient LMs.\nHardware co-design: on-device inference optimization, NPU/accelerator-aware design, memory-bandwidth-bound serving, and energy-budgeted decoding.\nTraining for cooperation: fine-tuning SLMs for tool use, planning, multi-step reasoning, and small-large model handoff in heterogeneous agent stacks.\nEvaluation and benchmarks: task-success-per-watt, latency- and memory-aware leaderboards, and reproducible on-device evaluation harnesses.\nApplications and safety: privacy-preserving local processing, federated learning, and deployment case studies across mobile assistants, robotics, healthcare, automotive, and financial services, with associated safety, robustness, and provenance considerations.\n\nCall for Papers\nWe invite original work in progress on small language models for agentic systems.\n\nSubmission format\n- Extended abstracts: 4 pages plus references\n- Optional full papers: 8 pages plus references\n- NeurIPS workshop template\n- Double-blind review through OpenReview\n- Three reviewers per submission\n\nScope and publication\n- Original work in progress\n- Not under review at or accepted to the NeurIPS 2026 main program\n- Not previously published at a major ML or AI venue\n- Accepted papers hosted on OpenReview with author opt-in\n- Non-archival and not included in the NeurIPS proceedings\n\nTopics of interest\n- SLM architectures, training and inference\n- Agentic reasoning, planning and tool use\n- Hardware-aware and on-device deployment\n- Evaluation, benchmarks and efficiency\n- Privacy, safety, robustness and applications\n\nImportant Dates\nCall for Papers released: July 18, 2026\nSubmission portal opens: July 25, 2026\nPaper submission deadline: August 22, 2026 (AoE)\nReviewing period: August 30-September 19, 2026\nAcceptance notification: September 22, 2026\nCamera-ready deadline: October 13, 2026\nWorkshop: December 12 or 13, 2026\n\nEdge Agent Efficiency Challenge\nThe challenge focuses on running a multi-step agent task on consumer-class hardware, such as a laptop GPU, mobile NPU, or single-board computer, under fixed memory, latency, and energy budgets. Entries will be evaluated using a cost-adjusted task-success metric and will release reusable model checkpoints and inference recipes. Challenge winners will present their systems during the workshop's live-demo session.\n\nPolicies\nNon-archival workshop: Accepted workshop submissions are non-archival and may be submitted elsewhere after the workshop, subject to the policies of the other venue.\nParticipation and inclusion: The program includes mentorship for junior researchers, a dedicated lightning-talk track, outreach through affinity communities, and an inclusive workshop environment for participants across backgrounds and career stages.\n\nFor questions, contact the organizers at neurips.slmagents.2026@gmail.com." }, { "key": "TCCML", "title": "Tackling Climate Change with Machine Learning: workshop at NeurIPS 2026", "subtitle": "TCCML @ NeurIPS 2026", "summary": "This workshop highlights work demonstrating that machine learning can be an invaluable tool in climate change mitigation, adaptation, and climate science, bringing together ML researchers and experts in climate-relevant fields. The 2026 theme, 'Fostering Ground-Up Innovation in AI and Climate,' encourages application-driven ML research, community-led data initiatives, open-source tools, and interdisciplinary collaboration.", "cfp_full": "NeurIPS 2026 Workshop: Tackling Climate Change with Machine Learning\n\nAbout\nMany in the ML community wish to take action on climate change, but are unsure how to have the most impact. This workshop will highlight work that demonstrates that, while ML is no silver bullet, it can be an invaluable tool in mitigation (reducing greenhouse gas emissions), adaptation (helping society respond to the effects of climate change), and climate science (improving understanding of climate change).\nClimate change is a complex problem for which action takes many forms, from advancing theory to deploying new technology. Many of these actions represent high-impact opportunities for real-world change, and simultaneously pose interesting academic research problems.\nThis workshop is part of a series that aims to bring together those applying ML to climate change challenges and facilitate cross-pollination between ML researchers and experts in climate-relevant fields.\nThe theme of this workshop, \"Fostering Ground-Up Innovation in AI and Climate,\" invites submissions that explore the strengths of diverse machine learning approaches in climate-related contexts. We particularly encourage work that develops ground-up approaches to AI for climate action, including ML research driven by an application-specific need or constraints, community-led data initiatives, open source tools, and interdisciplinary collaborations. Our goal is to foster a discussion on an alternative path for how domain-specific ML innovation is conducted, which may involve smaller or hybrid models, community-generated datasets, and consider practical tensions around resources, evaluation, scalability, and impact.\nThe main workshop will take place on December 11/12, 2026.\n\nImportant Dates\nMentorship program application deadline: August 1, 2026 extended to August 5, 2026, 23:59 AoE\nRecommended date to create your OpenReview account: August 15, 2026 (account creation can take up to 2 weeks)\nAbstract submission deadline: August 22, 2026 extended to August 29, 2026, 23:59 AoE\nSubmission deadline for workshop contributions: August 29, 2026, 23:59 AoE\nDecisions released: September 29, 2026, 23:59 AoE\nWorkshop dates: December 11/12, 2026\n\nCall for Submissions\nIMPORTANT: Starting with this year's workshop, we will be using OpenReview for managing submissions and conducting the review process. OpenReview account creation is not instant, and can take up to 2 weeks. We recommend going through the account creation process early to ensure you can submit on time.\nWe invite submissions of short papers, proposals, or tutorial notebooks using machine learning to address problems in climate mitigation, adaptation, or science, including but not limited to the following topics:\n- Agriculture and food\n- Behavioural and social science\n- Buildings\n- Carbon capture and sequestration\n- Cities and urban planning\n- Climate finance and economics\n- Climate justice\n- Climate science and climate modeling\n- Disaster management and relief\n- Earth observations and monitoring\n- Earth science\n- Ecosystems and biodiversity\n- Extreme weather\n- Forestry and other land use\n- Health\n- Heavy industry and manufacturing\n- Local and indigenous knowledge systems\n- Materials science and discovery\n- Oceans and marine systems\n- Power and energy systems\n- Public policy\n- Societal adaptation and resilience\n- Supply chains\n- Transportation\nAll machine learning techniques are welcome, from kernel methods to deep learning. Each submission should make clear why the application has (or could have) a pathway to positive impacts regarding climate change. We highly encourage submissions which make their data and code publicly available. Accepted submissions will be invited to give poster presentations, of which some will be selected for spotlight talks.\nThe workshop does not publish proceedings, and submissions are non-archival. Submission to this workshop does not preclude future publication. Previously published work may be submitted under certain circumstances (see the FAQ).\nAll papers, proposals, and tutorial submissions must be through the submission website. Please note that the submission link is not yet active; we expect it to open by around August 10, 2026.\nSubmissions will be reviewed double-blind; do your best to anonymize your submission, and do not include identifying information for authors in the PDF. Authors are required to use the workshop style template (based on the NeurIPS style files), available for LaTeX.\nBesides adhering to the NeurIPS 2026 LLM Policy on the Use of Large Language Models, this workshop requires that no CCAI website materials, including its calls for papers, should be shared with LLMs in whole or in part during the paper-writing process. Please keep in mind that the CCAI content mentioned above is copyrighted material.\n\nSubmission Tracks\nThere are three tracks for submissions: (i) Papers, (ii) Proposals, (iii) Tutorials. Submissions are limited to four pages for the Papers track, and three pages for the Proposals track, in PDF format. References do not count towards this total. Supplementary appendices are allowed but will be read at the discretion of the reviewers. Tutorial submissions are in executable notebook format that follows CCAI's NeurIPS 2026 Tutorial Template. All submissions must explain why the proposed work has (or could have) positive impacts regarding climate change.\n\nPAPERS Track\nWork that is in progress, published, and/or deployed.\nSubmissions for the Papers track should describe projects relevant to climate change that involve machine learning. These may include (but are not limited to) academic research; deployed results from startups, industry, public institutions, etc. and climate-relevant datasets.\nSubmissions should provide experimental or theoretical validation of the method presented, as well as specifying what gap the method fills. Authors should clearly illustrate a pathway to climate impact, i.e., identify the way in which this work fits into broader efforts to address climate change. Algorithms need not be novel from a machine learning perspective if they are applied in a novel setting. Details of methodology need not be revealed if they are proprietary, though transparency is highly encouraged.\nSubmissions creating novel datasets are welcome. Datasets should be properly documented with regards to their provenance and contents and designed to permit machine learning research (e.g. formatted with clear benchmarks for evaluation).\n\nPROPOSALS Track\nEarly-stage work and detailed descriptions of ideas for future work.\nSubmissions for the Proposals track should describe detailed ideas for how machine learning can be used to solve climate-relevant problems. While less constrained than the Papers track, Proposals will be subject to a very high standard of review. Ideas should be justified as extensively as possible, including motivation for why the problem being solved is important in tackling climate change, a discussion of why current methods are inadequate, an explanation of the proposed method, and a discussion of the pathway to climate impact. Preliminary results are optional.\n\nTUTORIALS Track\nInteractive notebooks for insightful step-by-step walkthroughs\nSubmissions for the Tutorials track should demonstrate the use of machine learning methods and tools (e.g., libraries, packages, services, datasets, or frameworks) to address climate-relevant challenges.\nTutorial submissions should include a clear and concise description of user learning outcomes. Tutorial submissions will be reviewed based on their potential impact, pedagogical value, and usability by the climate and AI research community. Notebooks will undergo two review cycles to ensure high quality submissions. In the first round, Round I, reviews will be made upon the initial (80% complete) submission (due August 29, 2026) and the second round, Round II, of reviews will be made upon the complete (100%) submission (due October 15, 2026) which should address and revise content based on feedback from Round I. Tutorials should be in the form of executable notebooks that follow CCAI's NeurIPS 2026 Tutorial Template. Authors must submit the notebook along with an accompanying requirements.txt file to ensure users and reviewers are able to run the tutorial in a self-contained runnable environment both locally and remotely (e.g., Python virtual environments, Colab).", "cfp_status": "published", "topics": [ "Agriculture and food", "Behavioural and social science", "Buildings", "Carbon capture and sequestration", "Cities and urban planning", "Climate finance and economics", "Climate justice", "Climate science and climate modeling", "Disaster management and relief", "Earth observations and monitoring", "Earth science", "Ecosystems and biodiversity", "Extreme weather", "Forestry and other land use", "Health", "Heavy industry and manufacturing", "Local and indigenous knowledge systems", "Materials science and discovery", "Oceans and marine systems", "Power and energy systems", "Public policy", "Societal adaptation and resilience", "Supply chains", "Transportation" ], "important_dates": [ { "label": "Mentorship program application deadline (extended)", "date": "2026-08-05" }, { "label": "Recommended OpenReview account creation date", "date": "2026-08-15" }, { "label": "Abstract submission deadline (extended)", "date": "2026-08-29" }, { "label": "Submission deadline for workshop contributions", "date": "2026-08-29" }, { "label": "Decisions released", "date": "2026-09-29" }, { "label": "Workshop dates", "date": "December 11/12, 2026" } ], "organizers": [ "Utkarsha Agwan (Tyba Energy)", "John Duncan (University of Western Australia)", "Kim Bente (University of Tasmania)", "Rajanie Prabha (Stanford University)", "Jonathan Richetti (CSIRO)", "Chen Chen (Centre for Climate Research Singapore)", "Yuanyuan Shi (University of California, San Diego)", "David Rolnick (McGill University and Mila - Quebec AI Institute)" ], "speakers": [], "host_url": "https://www.climatechange.ai/events/neurips2026", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/TCCML", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/TCCML", "location": "Sydney, Australia", "city": "Sydney", "workshop_date": "2026-12-10", "contact": "climatechangeai.neurips2026@gmail.com", "tracks": [ { "key": "TCCML", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/TCCML", "submission_dates_raw": "" } ], "group": "science", "group_label": "AI for Science", "corpus": "Tackling Climate Change with Machine Learning: workshop at NeurIPS 2026 TCCML @ NeurIPS 2026 This workshop highlights work demonstrating that machine learning can be an invaluable tool in climate change mitigation, adaptation, and climate science, bringing together ML researchers and experts in climate-relevant fields. The 2026 theme, 'Fostering Ground-Up Innovation in AI and Climate,' encourages application-driven ML research, community-led data initiatives, open-source tools, and interdisciplinary collaboration. Agriculture and food Behavioural and social science Buildings Carbon capture and sequestration Cities and urban planning Climate finance and economics Climate justice Climate science and climate modeling Disaster management and relief Earth observations and monitoring Earth science Ecosystems and biodiversity Extreme weather Forestry and other land use Health Heavy industry and manufacturing Local and indigenous knowledge systems Materials science and discovery Oceans and marine systems Power and energy systems Public policy Societal adaptation and resilience Supply chains Transportation NeurIPS 2026 Workshop: Tackling Climate Change with Machine Learning\n\nAbout\nMany in the ML community wish to take action on climate change, but are unsure how to have the most impact. This workshop will highlight work that demonstrates that, while ML is no silver bullet, it can be an invaluable tool in mitigation (reducing greenhouse gas emissions), adaptation (helping society respond to the effects of climate change), and climate science (improving understanding of climate change).\nClimate change is a complex problem for which action takes many forms, from advancing theory to deploying new technology. Many of these actions represent high-impact opportunities for real-world change, and simultaneously pose interesting academic research problems.\nThis workshop is part of a series that aims to bring together those applying ML to climate change challenges and facilitate cross-pollination between ML researchers and experts in climate-relevant fields.\nThe theme of this workshop, \"Fostering Ground-Up Innovation in AI and Climate,\" invites submissions that explore the strengths of diverse machine learning approaches in climate-related contexts. We particularly encourage work that develops ground-up approaches to AI for climate action, including ML research driven by an application-specific need or constraints, community-led data initiatives, open source tools, and interdisciplinary collaborations. Our goal is to foster a discussion on an alternative path for how domain-specific ML innovation is conducted, which may involve smaller or hybrid models, community-generated datasets, and consider practical tensions around resources, evaluation, scalability, and impact.\nThe main workshop will take place on December 11/12, 2026.\n\nImportant Dates\nMentorship program application deadline: August 1, 2026 extended to August 5, 2026, 23:59 AoE\nRecommended date to create your OpenReview account: August 15, 2026 (account creation can take up to 2 weeks)\nAbstract submission deadline: August 22, 2026 extended to August 29, 2026, 23:59 AoE\nSubmission deadline for workshop contributions: August 29, 2026, 23:59 AoE\nDecisions released: September 29, 2026, 23:59 AoE\nWorkshop dates: December 11/12, 2026\n\nCall for Submissions\nIMPORTANT: Starting with this year's workshop, we will be using OpenReview for managing submissions and conducting the review process. OpenReview account creation is not instant, and can take up to 2 weeks. We recommend going through the account creation process early to ensure you can submit on time.\nWe invite submissions of short papers, proposals, or tutorial notebooks using machine learning to address problems in climate mitigation, adaptation, or science, including but not limited to the following topics:\n- Agriculture and food\n- Behavioural and social science\n- Buildings\n- Carbon capture and sequestration\n- Cities and urban planning\n- Climate finance and economics\n- Climate justice\n- Climate science and climate modeling\n- Disaster management and relief\n- Earth observations and monitoring\n- Earth science\n- Ecosystems and biodiversity\n- Extreme weather\n- Forestry and other land use\n- Health\n- Heavy industry and manufacturing\n- Local and indigenous knowledge systems\n- Materials science and discovery\n- Oceans and marine systems\n- Power and energy systems\n- Public policy\n- Societal adaptation and resilience\n- Supply chains\n- Transportation\nAll machine learning techniques are welcome, from kernel methods to deep learning. Each submission should make clear why the application has (or could have) a pathway to positive impacts regarding climate change. We highly encourage submissions which make their data and code publicly available. Accepted submissions will be invited to give poster presentations, of which some will be selected for spotlight talks.\nThe workshop does not publish proceedings, and submissions are non-archival. Submission to this workshop does not preclude future publication. Previously published work may be submitted under certain circumstances (see the FAQ).\nAll papers, proposals, and tutorial submissions must be through the submission website. Please note that the submission link is not yet active; we expect it to open by around August 10, 2026.\nSubmissions will be reviewed double-blind; do your best to anonymize your submission, and do not include identifying information for authors in the PDF. Authors are required to use the workshop style template (based on the NeurIPS style files), available for LaTeX.\nBesides adhering to the NeurIPS 2026 LLM Policy on the Use of Large Language Models, this workshop requires that no CCAI website materials, including its calls for papers, should be shared with LLMs in whole or in part during the paper-writing process. Please keep in mind that the CCAI content mentioned above is copyrighted material.\n\nSubmission Tracks\nThere are three tracks for submissions: (i) Papers, (ii) Proposals, (iii) Tutorials. Submissions are limited to four pages for the Papers track, and three pages for the Proposals track, in PDF format. References do not count towards this total. Supplementary appendices are allowed but will be read at the discretion of the reviewers. Tutorial submissions are in executable notebook format that follows CCAI's NeurIPS 2026 Tutorial Template. All submissions must explain why the proposed work has (or could have) positive impacts regarding climate change.\n\nPAPERS Track\nWork that is in progress, published, and/or deployed.\nSubmissions for the Papers track should describe projects relevant to climate change that involve machine learning. These may include (but are not limited to) academic research; deployed results from startups, industry, public institutions, etc. and climate-relevant datasets.\nSubmissions should provide experimental or theoretical validation of the method presented, as well as specifying what gap the method fills. Authors should clearly illustrate a pathway to climate impact, i.e., identify the way in which this work fits into broader efforts to address climate change. Algorithms need not be novel from a machine learning perspective if they are applied in a novel setting. Details of methodology need not be revealed if they are proprietary, though transparency is highly encouraged.\nSubmissions creating novel datasets are welcome. Datasets should be properly documented with regards to their provenance and contents and designed to permit machine learning research (e.g. formatted with clear benchmarks for evaluation).\n\nPROPOSALS Track\nEarly-stage work and detailed descriptions of ideas for future work.\nSubmissions for the Proposals track should describe detailed ideas for how machine learning can be used to solve climate-relevant problems. While less constrained than the Papers track, Proposals will be subject to a very high standard of review. Ideas should be justified as extensively as possible, including motivation for why the problem being solved is important in tackling climate change, a discussion of why current methods are inadequate, an explanation of the proposed method, and a discussion of the pathway to climate impact. Preliminary results are optional.\n\nTUTORIALS Track\nInteractive notebooks for insightful step-by-step walkthroughs\nSubmissions for the Tutorials track should demonstrate the use of machine learning methods and tools (e.g., libraries, packages, services, datasets, or frameworks) to address climate-relevant challenges.\nTutorial submissions should include a clear and concise description of user learning outcomes. Tutorial submissions will be reviewed based on their potential impact, pedagogical value, and usability by the climate and AI research community. Notebooks will undergo two review cycles to ensure high quality submissions. In the first round, Round I, reviews will be made upon the initial (80% complete) submission (due August 29, 2026) and the second round, Round II, of reviews will be made upon the complete (100%) submission (due October 15, 2026) which should address and revise content based on feedback from Round I. Tutorials should be in the form of executable notebooks that follow CCAI's NeurIPS 2026 Tutorial Template. Authors must submit the notebook along with an accompanying requirements.txt file to ensure users and reviewers are able to run the tutorial in a self-contained runnable environment both locally and remotely (e.g., Python virtual environments, Colab)." }, { "key": "IAB", "title": "The 1st Workshop on Interpreting Agent Behavior (IAB) at NeurIPS 2026", "subtitle": "IAB @NeurIPS 2026 Workshop", "summary": "IAB works toward an interpretive science of agent behavior, turning agent runtime data (logs and interaction traces) into human understanding of what agents do and how humans work with them across three levels: the agent, the human, and their interaction. It bridges social science and HCI interpretation methods with AI's problem space of evaluation, governance, alignment, and responsible AI.", "cfp_full": "The First Workshop on Interpreting Agent Behavior\nHuman-Centered Interpretation for Understanding Agents, Humans, and Interaction\nNeurIPS 2026 · International Convention Centre, Sydney, Australia · December 11-12, 2026\n\nAbout\nAgents now run for hours, even days, to finish a single task. Along the way they plan, reason, use tools, and recover from errors. Every run leaves behind vast behavioral data: logs and interaction traces. Yet we still lack the vocabulary, methods, and tools to make sense of it at scale. Humans cannot read through thousands of log entries. They need patterns, summaries, and explanations, in other words interpretation. But we do not yet know how to generate it at scale. Our aim with IAB is to turn agent runtime data into human understanding of what agents do and how humans work with them.\nCommercial autonomous agents such as Claude and Codex now run for hours or even days to complete tasks, and along the way they show complex behavior: they plan, reason, use tools, recover from errors, coordinate with subagents, and communicate with users. We use the word behavior, as in the study of human behavior, for the full range of what an agent does during runtime. This behavior spans three levels: what agents do and how they do it, what humans do in response, and how the two work together through instructions and corrections. All three generate vast behavioral data such as execution logs and interaction traces. Yet existing approaches read this data largely for outcomes: benchmarks tell us whether an agent succeeds or fails, but not what it did or how it did it.\nUnderstanding what and how is what people actually need. It lets agent developers and model trainers debug failures, compare architectures, and filter training data; it lets agent users and deployment engineers watch production agents to understand safety, cost, and reliability risks. For agentic models, the trajectory is both the training data and what the reward scores. Interpreting it therefore sits inside the training loop, deciding which rollouts are safe to reinforce and flagging reward that reflects a verifier exploit rather than real skill. But the field still lacks the vocabulary, methods, and tools to describe and analyze agent behavior at scale. Humans cannot read through thousands of log entries; they need patterns, summaries, and explanations, in other words interpretation, and we do not yet know how to scale it.\nIAB works toward an interpretive science of agent behavior. It treats behavior across these three levels as the object of study and proceeds in two steps: first gathering the community to identify the problem space and emerging challenges, then bringing the broad set of methods that social scientists have developed, such as grounded theory, qualitative analysis, error analysis, corpus analysis, trace analysis, and red-teaming, to read meaning from this data, discover categories from it, and count them. IAB bridges two communities: social science and HCI contribute the interpretation methods, while AI contributes the problem space of evaluation, governance, alignment, and responsible AI.\n\nScope\nIAB studies agent behavior at three levels: the agent, the human, and their interaction. At each level we ask what happens and how.\nAgents - What do agents do, and how?\n- What do agents actually do during a run, from planning and reasoning to using tools, failing, and recovering?\n- What new or emergent behaviors show up in single- and multi-agent systems?\n- How are these behaviors shaped, by the model, the prompt, the harness, the skills, or the overall agent design?\nHumans - What do humans do in response, and how?\n- What do humans do while working with an agent, from writing prompts to verifying outputs and monitoring the run?\n- When should humans trust and rely on agents, and when should they not?\n- How do humans communicate intent, and step in to steer or correct an agent during runtime?\nInteraction - What happens when humans and agents interact, and how?\n- What patterns emerge in interaction traces, and how do they differ across tasks?\n- How do humans and agents build a shared understanding of the goal, and where does it break down?\n- How do they recover from misunderstandings as the work goes on?\n\nCall for Papers\nTopics\nWe call for non-archival submissions on understanding agent behavior. Work can address any level of our Scope (what agents do, what humans do in response, or how the two interact). Specific topics include, but are not limited to:\n- Datasets and resources: Data and resources for studying, debugging, and diagnosing agent behavior. Examples: Agent trajectory data on specific tasks; Human-agent conversational data and dialogue logs; Simulated or synthetic behavior data; Annotated interaction traces and trajectory data; Schemas for representing agent trajectories; Benchmarks and evaluations on specific behaviors; Discussion and positions, e.g., on how we should represent agent trajectories, or on how benchmarks and metrics can fail to capture agent behavior.\n- Empirical studies: Studies on how agents and people actually behave at runtime. Examples: Interpretations that benchmarks and metrics cannot reveal, e.g., qualitative analysis of agent trajectories; Quantitative and computational analysis of agent behavior, e.g., statistical patterns, clustering, or process mining over trajectories; Observational studies of agent behavior, in general or focused on specific behaviors like sycophancy, deception, or anthropomorphism; Case studies of individual agent runs; Automated, large-scale analysis that surfaces behavioral patterns across many runs; Identification of emergent behavior in single- and multi-agent systems; Behavior comparisons across models, prompts, or architectures; Behavior taxonomies for describing what happened inside an agent run; How agents behave in downstream scenarios, e.g., deployment, debugging, and diagnosis; Studies of how people work with agents, e.g., observations, interviews, and surveys.\n- Methods and tools: Approaches that help people describe, interpret, and oversee what agents do. Examples: Interaction and trajectory visualization tools that make agent behavior easier to understand; Human-centered theoretical perspectives on agent behavior, borrowing from disciplines such as social science, behavioral science; Theoretical vocabularies for describing what an agent does; Reusable behavior catalogs and qualitative coding codebooks for what an agent does; Grounded theory, thematic analysis, conversation analysis, and other qualitative methods applied to agent behavior; Automated methods for labeling, clustering, or summarizing agent behavior at scale, e.g., using LLM judges or embeddings; Metrics and probes that characterize agent behavior beyond task success; Discussion or position papers, e.g., on what cross-disciplinary methods we can use for human-centered interpretation of agent behaviors.\nWe also welcome negative results and methodological position papers. NeurIPS-rejected papers may be resubmitted with their reviews, and we will decide on acceptance ourselves. NeurIPS-accepted papers that want more visibility are also welcome to resubmit with their reviews.\n\nSubmission Format\nLong Papers: Up to 9 pages + references. For full empirical studies, datasets, or comprehensive analyses.\nShort Papers: Up to 4 pages + references. For position papers, tools, demos, preliminary findings, and negative results.\nReview Process: NeurIPS-style formatting, double-blind review, submission site: OpenReview.\nSubmission with NeurIPS reviews: If your paper was reviewed at NeurIPS 2026, you can submit it here together with the reviews and your response. This is a second route into the workshop. Due October 1, 2026, one week after NeurIPS decisions. The organizing team gives these papers a light review, so you can expect a decision within one or two weeks. Because the workshop is non-archival, presenting here does not affect where you send the paper elsewhere.\n\nImportant Dates\nSubmission opens: July 22, 2026\nSubmission deadline (without reviews): August 29, 2026\nReview period: August 31 - September 20, 2026\nDiscussion period: September 21-27, 2026\nDecision notification: September 29, 2026\n\"Submission with NeurIPS reviews\" deadline: October 1, 2026\n\"Submission with NeurIPS reviews\" notification: TBD\nCamera-ready deadline: November 20, 2026\nWorkshop date: December 11-12, 2026\n\nEthics and LLM Usage\nSubmissions must adhere to the NeurIPS main conference policies on the use of agents and large language models, as well as on research ethics, including obtaining appropriate IRB/institutional approval for studies involving human participants. Consult the NeurIPS 2026 Main Track Handbook for details.\nTravel grants/paper awards will be available.\n\nCompetition & Call for Competition Papers\nGLEE (Games in Language-based Economic Environments) is the official competition of the IAB Workshop at NeurIPS 2026. It evaluates AI agents in multi-turn bargaining, negotiation, and persuasion games, where success requires natural-language communication, strategic reasoning, adaptation to other players, and effective economic decision-making. Participants can build an autonomous agent that plays live through the GLEE API, or compete in the human track directly through the web interface. Agents and humans play online in a shared pool from August 1-29, 2026 (AoE), with a total prize pool of $6,000: $5,000 for the agent track and $1,000 for the human track. Participants may also submit a four-page competition paper describing their agent, approach, and findings. Accepted papers will be presented at the IAB Workshop at NeurIPS 2026, with a dedicated poster session and a Best Competition Paper Award. Please see more information here: https://glee-competition.com/", "cfp_status": "published", "topics": [ "Datasets and resources for studying, debugging, and diagnosing agent behavior", "Agent trajectory data, dialogue logs, and annotated interaction traces", "Schemas for representing agent trajectories", "Benchmarks and evaluations on specific behaviors", "Empirical and observational studies of agent behavior at runtime", "Qualitative and quantitative/computational analysis of agent trajectories", "Identification of emergent behavior in single- and multi-agent systems", "Behavior comparisons across models, prompts, or architectures", "Behavior taxonomies for describing agent runs", "Studies of how people work with agents (observations, interviews, surveys)", "Interaction and trajectory visualization tools", "Human-centered theoretical perspectives and vocabularies for agent behavior", "Grounded theory, thematic analysis, conversation analysis and qualitative methods", "Automated methods for labeling, clustering, or summarizing agent behavior at scale", "Metrics and probes beyond task success", "Negative results and methodological position papers" ], "important_dates": [ { "label": "Submission opens", "date": "2026-07-22" }, { "label": "Submission deadline (without reviews)", "date": "2026-08-29" }, { "label": "Review period", "date": "August 31 - September 20, 2026" }, { "label": "Discussion period", "date": "September 21-27, 2026" }, { "label": "Decision notification", "date": "2026-09-29" }, { "label": "\"Submission with NeurIPS reviews\" deadline", "date": "2026-10-01" }, { "label": "\"Submission with NeurIPS reviews\" notification", "date": "TBD" }, { "label": "Camera-ready deadline", "date": "2026-11-20" }, { "label": "Workshop date", "date": "December 11-12, 2026" } ], "organizers": [ "Jie (Sophia) Gao (Johns Hopkins University)", "Kaiser Sun (Johns Hopkins University)", "Teresa Yeo (Google DeepMind)", "Daniel Khashabi (Johns Hopkins University)", "Zhuoran Lu (University of Hong Kong)", "Boyuan Zheng (xAI)", "Katherine Van Koevering (Johns Hopkins University)", "Sijie Ji (Caltech)", "Jen-tse Huang (Johns Hopkins University)" ], "speakers": [ "Armando Solar-Lezama (MIT CSAIL)", "Diyi Yang (Stanford University)", "Been Kim (Google DeepMind)", "Marc-Alexandre Côté (Microsoft Research)", "Bowen Baker (OpenAI)", "Dinh Phung (Monash University)", "Kun Zhang (Carnegie Mellon University & MBZUAI)", "Nancy F. Chen (A*STAR Singapore)" ], "host_url": "https://iab-agents.github.io/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/IAB", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/IAB", "location": "Sydney, Australia", "city": "Sydney", "workshop_date": "2026-12-10", "contact": "gaojie056@gmail.com", "tracks": [ { "key": "IAB", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/IAB", "submission_dates_raw": "" }, { "key": "IAB_Competition_Paper_Track", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/IAB_Competition_Paper_Track", "submission_dates_raw": "" } ], "group": "agents", "group_label": "Agents & Agentic Systems", "corpus": "The 1st Workshop on Interpreting Agent Behavior (IAB) at NeurIPS 2026 IAB @NeurIPS 2026 Workshop IAB works toward an interpretive science of agent behavior, turning agent runtime data (logs and interaction traces) into human understanding of what agents do and how humans work with them across three levels: the agent, the human, and their interaction. It bridges social science and HCI interpretation methods with AI's problem space of evaluation, governance, alignment, and responsible AI. Datasets and resources for studying, debugging, and diagnosing agent behavior Agent trajectory data, dialogue logs, and annotated interaction traces Schemas for representing agent trajectories Benchmarks and evaluations on specific behaviors Empirical and observational studies of agent behavior at runtime Qualitative and quantitative/computational analysis of agent trajectories Identification of emergent behavior in single- and multi-agent systems Behavior comparisons across models, prompts, or architectures Behavior taxonomies for describing agent runs Studies of how people work with agents (observations, interviews, surveys) Interaction and trajectory visualization tools Human-centered theoretical perspectives and vocabularies for agent behavior Grounded theory, thematic analysis, conversation analysis and qualitative methods Automated methods for labeling, clustering, or summarizing agent behavior at scale Metrics and probes beyond task success Negative results and methodological position papers The First Workshop on Interpreting Agent Behavior\nHuman-Centered Interpretation for Understanding Agents, Humans, and Interaction\nNeurIPS 2026 · International Convention Centre, Sydney, Australia · December 11-12, 2026\n\nAbout\nAgents now run for hours, even days, to finish a single task. Along the way they plan, reason, use tools, and recover from errors. Every run leaves behind vast behavioral data: logs and interaction traces. Yet we still lack the vocabulary, methods, and tools to make sense of it at scale. Humans cannot read through thousands of log entries. They need patterns, summaries, and explanations, in other words interpretation. But we do not yet know how to generate it at scale. Our aim with IAB is to turn agent runtime data into human understanding of what agents do and how humans work with them.\nCommercial autonomous agents such as Claude and Codex now run for hours or even days to complete tasks, and along the way they show complex behavior: they plan, reason, use tools, recover from errors, coordinate with subagents, and communicate with users. We use the word behavior, as in the study of human behavior, for the full range of what an agent does during runtime. This behavior spans three levels: what agents do and how they do it, what humans do in response, and how the two work together through instructions and corrections. All three generate vast behavioral data such as execution logs and interaction traces. Yet existing approaches read this data largely for outcomes: benchmarks tell us whether an agent succeeds or fails, but not what it did or how it did it.\nUnderstanding what and how is what people actually need. It lets agent developers and model trainers debug failures, compare architectures, and filter training data; it lets agent users and deployment engineers watch production agents to understand safety, cost, and reliability risks. For agentic models, the trajectory is both the training data and what the reward scores. Interpreting it therefore sits inside the training loop, deciding which rollouts are safe to reinforce and flagging reward that reflects a verifier exploit rather than real skill. But the field still lacks the vocabulary, methods, and tools to describe and analyze agent behavior at scale. Humans cannot read through thousands of log entries; they need patterns, summaries, and explanations, in other words interpretation, and we do not yet know how to scale it.\nIAB works toward an interpretive science of agent behavior. It treats behavior across these three levels as the object of study and proceeds in two steps: first gathering the community to identify the problem space and emerging challenges, then bringing the broad set of methods that social scientists have developed, such as grounded theory, qualitative analysis, error analysis, corpus analysis, trace analysis, and red-teaming, to read meaning from this data, discover categories from it, and count them. IAB bridges two communities: social science and HCI contribute the interpretation methods, while AI contributes the problem space of evaluation, governance, alignment, and responsible AI.\n\nScope\nIAB studies agent behavior at three levels: the agent, the human, and their interaction. At each level we ask what happens and how.\nAgents - What do agents do, and how?\n- What do agents actually do during a run, from planning and reasoning to using tools, failing, and recovering?\n- What new or emergent behaviors show up in single- and multi-agent systems?\n- How are these behaviors shaped, by the model, the prompt, the harness, the skills, or the overall agent design?\nHumans - What do humans do in response, and how?\n- What do humans do while working with an agent, from writing prompts to verifying outputs and monitoring the run?\n- When should humans trust and rely on agents, and when should they not?\n- How do humans communicate intent, and step in to steer or correct an agent during runtime?\nInteraction - What happens when humans and agents interact, and how?\n- What patterns emerge in interaction traces, and how do they differ across tasks?\n- How do humans and agents build a shared understanding of the goal, and where does it break down?\n- How do they recover from misunderstandings as the work goes on?\n\nCall for Papers\nTopics\nWe call for non-archival submissions on understanding agent behavior. Work can address any level of our Scope (what agents do, what humans do in response, or how the two interact). Specific topics include, but are not limited to:\n- Datasets and resources: Data and resources for studying, debugging, and diagnosing agent behavior. Examples: Agent trajectory data on specific tasks; Human-agent conversational data and dialogue logs; Simulated or synthetic behavior data; Annotated interaction traces and trajectory data; Schemas for representing agent trajectories; Benchmarks and evaluations on specific behaviors; Discussion and positions, e.g., on how we should represent agent trajectories, or on how benchmarks and metrics can fail to capture agent behavior.\n- Empirical studies: Studies on how agents and people actually behave at runtime. Examples: Interpretations that benchmarks and metrics cannot reveal, e.g., qualitative analysis of agent trajectories; Quantitative and computational analysis of agent behavior, e.g., statistical patterns, clustering, or process mining over trajectories; Observational studies of agent behavior, in general or focused on specific behaviors like sycophancy, deception, or anthropomorphism; Case studies of individual agent runs; Automated, large-scale analysis that surfaces behavioral patterns across many runs; Identification of emergent behavior in single- and multi-agent systems; Behavior comparisons across models, prompts, or architectures; Behavior taxonomies for describing what happened inside an agent run; How agents behave in downstream scenarios, e.g., deployment, debugging, and diagnosis; Studies of how people work with agents, e.g., observations, interviews, and surveys.\n- Methods and tools: Approaches that help people describe, interpret, and oversee what agents do. Examples: Interaction and trajectory visualization tools that make agent behavior easier to understand; Human-centered theoretical perspectives on agent behavior, borrowing from disciplines such as social science, behavioral science; Theoretical vocabularies for describing what an agent does; Reusable behavior catalogs and qualitative coding codebooks for what an agent does; Grounded theory, thematic analysis, conversation analysis, and other qualitative methods applied to agent behavior; Automated methods for labeling, clustering, or summarizing agent behavior at scale, e.g., using LLM judges or embeddings; Metrics and probes that characterize agent behavior beyond task success; Discussion or position papers, e.g., on what cross-disciplinary methods we can use for human-centered interpretation of agent behaviors.\nWe also welcome negative results and methodological position papers. NeurIPS-rejected papers may be resubmitted with their reviews, and we will decide on acceptance ourselves. NeurIPS-accepted papers that want more visibility are also welcome to resubmit with their reviews.\n\nSubmission Format\nLong Papers: Up to 9 pages + references. For full empirical studies, datasets, or comprehensive analyses.\nShort Papers: Up to 4 pages + references. For position papers, tools, demos, preliminary findings, and negative results.\nReview Process: NeurIPS-style formatting, double-blind review, submission site: OpenReview.\nSubmission with NeurIPS reviews: If your paper was reviewed at NeurIPS 2026, you can submit it here together with the reviews and your response. This is a second route into the workshop. Due October 1, 2026, one week after NeurIPS decisions. The organizing team gives these papers a light review, so you can expect a decision within one or two weeks. Because the workshop is non-archival, presenting here does not affect where you send the paper elsewhere.\n\nImportant Dates\nSubmission opens: July 22, 2026\nSubmission deadline (without reviews): August 29, 2026\nReview period: August 31 - September 20, 2026\nDiscussion period: September 21-27, 2026\nDecision notification: September 29, 2026\n\"Submission with NeurIPS reviews\" deadline: October 1, 2026\n\"Submission with NeurIPS reviews\" notification: TBD\nCamera-ready deadline: November 20, 2026\nWorkshop date: December 11-12, 2026\n\nEthics and LLM Usage\nSubmissions must adhere to the NeurIPS main conference policies on the use of agents and large language models, as well as on research ethics, including obtaining appropriate IRB/institutional approval for studies involving human participants. Consult the NeurIPS 2026 Main Track Handbook for details.\nTravel grants/paper awards will be available.\n\nCompetition & Call for Competition Papers\nGLEE (Games in Language-based Economic Environments) is the official competition of the IAB Workshop at NeurIPS 2026. It evaluates AI agents in multi-turn bargaining, negotiation, and persuasion games, where success requires natural-language communication, strategic reasoning, adaptation to other players, and effective economic decision-making. Participants can build an autonomous agent that plays live through the GLEE API, or compete in the human track directly through the web interface. Agents and humans play online in a shared pool from August 1-29, 2026 (AoE), with a total prize pool of $6,000: $5,000 for the agent track and $1,000 for the human track. Participants may also submit a four-page competition paper describing their agent, approach, and findings. Accepted papers will be presented at the IAB Workshop at NeurIPS 2026, with a dedicated poster session and a Best Competition Paper Award. Please see more information here: https://glee-competition.com/" }, { "key": "PhysWorldAI", "title": "The 1st Workshop on Physical World AI: Geometry, Characteristics, and Multimodal Sensing (NeurIPS 2026)", "subtitle": "PhysWorldAI - Beyond pixels: understanding the physical world", "summary": "PhysWorldAI is the first workshop on physical world AI, centered on how AI systems can perceive, represent, and reason about the physical world beyond appearance, organized around three coupled pillars: physical geometry, physical characteristics, and physical sensors.", "cfp_full": "PhysWorldAI - Physical World AI: Geometry, Characteristics, and Multimodal Sensing\n1st Workshop on Physical World AI\nBeyond pixels: understanding the physical world through geometry, characteristics, and multimodal sensing.\n\nOverview\n\nPhysWorldAI is the first workshop on physical world AI, centered on the question of how AI systems can perceive, represent, and reason about the physical world beyond appearance.\n\nThe workshop program is organized around three coupled pillars: physical geometry, physical characteristics, and physical sensors. It brings together computer vision, robotics, graphics, multimodal learning, haptics, audio, and physics-based simulation communities before benchmarks and protocols fragment across separate venues.\n\nThe half-day program includes moderated Q&A, contributed-paper spotlights, posters, live demos, and a synthesis panel across geometry, characteristics, sensing, and embodied AI in Atlanta, Georgia.\n\nTopics - Geometry, characteristics, sensors, and cross-cutting physical AI\n- Physical Geometry: 3D/4D reconstruction, articulated and deformable scene understanding, geometry-aware world models, and physically grounded view synthesis.\n- Physical Characteristics: Material and physical property estimation, mass, friction, stiffness, elasticity, deformability, affordances, contact-rich interaction, differentiable simulation, and generative models of physical dynamics.\n- Physical Sensors: Multimodal sensing and fusion with tactile, force/torque, proprioceptive, RF, audio, depth, IMU, and event-based signals.\n- Cross-cutting: Embodied world models, robot manipulation, sim-to-real transfer, multimodal simulators, benchmarks, datasets, evaluation protocols, and responsible deployment.\n\nSponsorship and Awards: Sponsor support is allocated to the Overall Best Paper Award, Best Student Paper Award, student travel grants for underrepresented attendees, and on-site logistics.\n\nKey Dates\n- Submission deadline: August 29, 2026\n- Notification: September 26, 2026\n- Camera-ready: October 26, 2026\n- Workshop: December 8-13, 2026 (Atlanta, Georgia)", "cfp_status": "published", "topics": [ "Physical Geometry: 3D/4D reconstruction, articulated and deformable scene understanding, geometry-aware world models, physically grounded view synthesis", "Physical Characteristics: material and physical property estimation, mass, friction, stiffness, elasticity, deformability, affordances, contact-rich interaction, differentiable simulation, generative models of physical dynamics", "Physical Sensors: multimodal sensing and fusion with tactile, force/torque, proprioceptive, RF, audio, depth, IMU, and event-based signals", "Cross-cutting: embodied world models, robot manipulation, sim-to-real transfer, multimodal simulators, benchmarks, datasets, evaluation protocols, responsible deployment" ], "important_dates": [ { "label": "Submission deadline", "date": "2026-08-29" }, { "label": "Notification", "date": "2026-09-26" }, { "label": "Camera-ready", "date": "2026-10-26" }, { "label": "Workshop", "date": "December 8-13, 2026" } ], "organizers": [ "Kaichen Zhou (MIT / Harvard University) - Contact person", "Ruojin Cai (Harvard University)", "Jianqing Zheng (University of Oxford)", "Congyue Deng (MIT)", "Amir Jamaludin (University of Oxford)", "Yining Hong (Stanford University)", "Shangzhe Wu (University of Cambridge)", "Mengyu Wang (Harvard Medical School)", "Rao Fu (Brown University)", "Fangneng Zhan (Hong Kong University of Science and Technology)", "Wei Dai (MIT)", "Tiange Xiang (Stanford University / MIT)", "Zihan Wang (Abaka AI / 2077AI)", "Xinhai Chang (Junior Organizer, Peking University)", "Yuzhen Chen (Junior Organizer, Harvard Medical School)", "Zeyang Bai (Junior Organizer, WorldMind Lab)" ], "speakers": [ "Anima Anandkumar (Caltech)", "Noah Snavely (Cornell University / Cornell Tech / Google DeepMind)", "Chelsea Finn (Stanford University)", "William T. Freeman (MIT)" ], "host_url": "https://physworld-org.github.io/physworld.github.io/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/PhysWorldAI", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/PhysWorldAI", "location": "Atlanta, Georgia, United States", "city": "Atlanta", "workshop_date": "", "contact": "physicalworldai@hotmail.com", "tracks": [ { "key": "PhysWorldAI", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/PhysWorldAI", "submission_dates_raw": "Submission Start: Aug 01 2026 11:59PM UTC-0, Submission Deadline: Aug 29 2026 11:59PM UTC-0" } ], "group": "science", "group_label": "AI for Science", "corpus": "The 1st Workshop on Physical World AI: Geometry, Characteristics, and Multimodal Sensing (NeurIPS 2026) PhysWorldAI - Beyond pixels: understanding the physical world PhysWorldAI is the first workshop on physical world AI, centered on how AI systems can perceive, represent, and reason about the physical world beyond appearance, organized around three coupled pillars: physical geometry, physical characteristics, and physical sensors. Physical Geometry: 3D/4D reconstruction, articulated and deformable scene understanding, geometry-aware world models, physically grounded view synthesis Physical Characteristics: material and physical property estimation, mass, friction, stiffness, elasticity, deformability, affordances, contact-rich interaction, differentiable simulation, generative models of physical dynamics Physical Sensors: multimodal sensing and fusion with tactile, force/torque, proprioceptive, RF, audio, depth, IMU, and event-based signals Cross-cutting: embodied world models, robot manipulation, sim-to-real transfer, multimodal simulators, benchmarks, datasets, evaluation protocols, responsible deployment PhysWorldAI - Physical World AI: Geometry, Characteristics, and Multimodal Sensing\n1st Workshop on Physical World AI\nBeyond pixels: understanding the physical world through geometry, characteristics, and multimodal sensing.\n\nOverview\n\nPhysWorldAI is the first workshop on physical world AI, centered on the question of how AI systems can perceive, represent, and reason about the physical world beyond appearance.\n\nThe workshop program is organized around three coupled pillars: physical geometry, physical characteristics, and physical sensors. It brings together computer vision, robotics, graphics, multimodal learning, haptics, audio, and physics-based simulation communities before benchmarks and protocols fragment across separate venues.\n\nThe half-day program includes moderated Q&A, contributed-paper spotlights, posters, live demos, and a synthesis panel across geometry, characteristics, sensing, and embodied AI in Atlanta, Georgia.\n\nTopics - Geometry, characteristics, sensors, and cross-cutting physical AI\n- Physical Geometry: 3D/4D reconstruction, articulated and deformable scene understanding, geometry-aware world models, and physically grounded view synthesis.\n- Physical Characteristics: Material and physical property estimation, mass, friction, stiffness, elasticity, deformability, affordances, contact-rich interaction, differentiable simulation, and generative models of physical dynamics.\n- Physical Sensors: Multimodal sensing and fusion with tactile, force/torque, proprioceptive, RF, audio, depth, IMU, and event-based signals.\n- Cross-cutting: Embodied world models, robot manipulation, sim-to-real transfer, multimodal simulators, benchmarks, datasets, evaluation protocols, and responsible deployment.\n\nSponsorship and Awards: Sponsor support is allocated to the Overall Best Paper Award, Best Student Paper Award, student travel grants for underrepresented attendees, and on-site logistics.\n\nKey Dates\n- Submission deadline: August 29, 2026\n- Notification: September 26, 2026\n- Camera-ready: October 26, 2026\n- Workshop: December 8-13, 2026 (Atlanta, Georgia)" }, { "key": "Simbiochem", "title": "The 2nd Workshop on Simulations for Biology and Chemistry", "subtitle": "Simbiochem 2026", "summary": "SIMBIOCHEM II is a NeurIPS 2026 workshop on Machine Learning for Simulations in Biology & Chemistry that brings together ML, computational chemistry, biophysics and materials science to fuse simulation, generative models and agentic AI into physics-aligned systems that learn from reality.", "cfp_full": "SIMBIOCHEM II — Machine Learning for Simulations in Biology & Chemistry\nThe 2nd SIMBIOCHEM Workshop\nNeurIPS 2026 Workshop · Sydney, Australia · Dec 11 or 12, 2026\nFusing molecular simulation, generative models and agentic AI into physics-aligned systems that learn from reality.\n\nThe workshop\nSimulation as the substrate for scientific AI\nThe next step beyond structure prediction is motion — trajectories, ensembles, kinetics and rare events. SIMBIOCHEM II brings together machine learning, computational chemistry, biophysics and materials science to make molecular AI faster, and grounded in physical rigour.\n\nThe guiding question: \"How can simulation, generative models, and agentic AI be fused into physics-aligned systems that learn from reality and accelerate biological and chemical discovery?\"\n\nThe workshop bridges machine learning with physical simulation, integrating scalable ML with rigorous physical simulation to make methods faster, and more accurate, reliable and grounded in science.\n\nTopics & themes\n- Learned potentials & force fields\n- Differentiable & enhanced molecular dynamics\n- Conformational ensembles, kinetics & rare events\n- Molecular foundation models & scientific post-training\n- Physical alignment: calibrated uncertainty & free energies\n- Agentic ML: tool-calling MD/QM, active learning, closed-loop discovery\n- Physically-grounded architectures (symmetries, conservation laws)\n- Differentiable simulation & inverse design\n- Next-generation learned potentials\n- Data efficiency & scalability, foundation models, agentic ML\n- Standardisation & reproducibility: benchmarks, datasets, metrics\n\nWho Can Submit\nA broad range of contributions across ML, computational chemistry, biophysics and materials science are welcome, particularly work showing how ML can accelerate physical simulation, and how simulation can inform and strengthen ML. Community: ML Researchers, Computational Chemistry, Biology, Materials Science, Biophysics, Life Sciences.\n\nReview Criteria\nSubmissions are evaluated on: Novelty (a new idea or genuinely new angle); Impact (does it matter to the field?); Correctness (are claims supported by evidence?); Clarity (can readers follow the method?).\n\nSubmission Format & Page Limits\nNon-archival short papers (5–8 pages) and abstracts, double-blind. Papers: 5–8 pages, excluding references, appendices and data-availability statements. References and appendices may be unlimited. Reviewers reserve the right to stop reading beyond 8 pages; extra content is at the author's risk. Six spotlight papers are selected for 10-minute talks; others present posters (24\" W × 36\" H, portrait orientation). Best-paper awards are given from accepted submissions.\n\nAnonymization Requirements\nSubmissions must be fully anonymised as per the template instructions. Do not include author names, affiliations or acknowledgements. Authors must remove any identifying links — GitHub repositories, personal websites, and images/figures (or their metadata) that could reveal your identity. Failing to submit an anonymised paper will cause desk rejection.\n\nConflict of Interest Policy\nThis year, NeurIPS requires disclosure of conflicts of interest. Authors must declare relationships with any organising or advisory committee member, including: same institution (current or past 3 years); PhD advisor/advisee relationships (any time); current or recent collaboration/co-authorship (past 3 years); family or close personal relationships; shared grants, funding, or acknowledgements; any other fairness-compromising relationship. Submissions from organisers, and from those with a personal COI to an organiser (their students, postdocs, close collaborators, family or close personal relationships), are not eligible.\n\nNon-archival policy\nThe workshop is non-archival, with no proceedings and no rebuttal phase. Camera-ready accepted papers are collated into a proceedings page on the website (authors may opt out). Authors may submit work previously presented at other workshops or conferences, or currently under review elsewhere.\n\nSubmission Platform\nSubmit via OpenReview at https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/Simbiochem\n\nKey Dates\n- Submission Deadline: August 29, 2026, 11:59 PM UTC\n- Author Notification: September 29, 2026 (AoE)\n- Workshop Day: Dec 11 or 12, 2026 (TBC by NeurIPS)\n- Camera-ready (accepted only): Date announced with decisions", "cfp_status": "published", "topics": [ "Learned potentials & force fields", "Differentiable & enhanced molecular dynamics", "Conformational ensembles, kinetics & rare events", "Molecular foundation models & scientific post-training", "Physical alignment: calibrated uncertainty & free energies", "Agentic ML: tool-calling MD/QM, active learning, closed-loop discovery", "Physically-grounded architectures (symmetries, conservation laws)", "Differentiable simulation & inverse design", "Data efficiency & scalability, foundation models", "Standardisation & reproducibility: benchmarks, datasets, metrics" ], "important_dates": [ { "label": "Submission Deadline", "date": "2026-08-29" }, { "label": "Author Notification", "date": "2026-09-29" }, { "label": "Workshop Day", "date": "2026-12-11 or 2026-12-12" } ], "organizers": [ "Bruno Trentini (University of Oxford · NVIDIA)", "Emine Kucukbenli (NVIDIA)", "Jigyasa Nigam (MIT)", "Max Secor (Novo Nordisk)", "Ole Winther (University of Copenhagen · DTU · Raffle.ai)", "Runzhong Wang (MIT, Coley Lab)" ], "speakers": [ "Frank Noé (Microsoft Research AI for Science · FU Berlin)", "Yu-Shan Lin (Tufts University)", "Heather J. Kulik (MIT)", "Ai Niitsu (RIKEN IMS)", "John Chodera (Sloan Kettering Institute · MSKCC)" ], "host_url": "https://www.simbiochem.com", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/Simbiochem", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/Simbiochem", "location": "Sydney, Australia", "city": "Sydney", "workshop_date": "", "contact": "workshop@simbiochem.com", "tracks": [ { "key": "Simbiochem", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/Simbiochem", "submission_dates_raw": "Submission Deadline: Aug 29 2026 11:59PM UTC-0" } ], "group": "science", "group_label": "AI for Science", "corpus": "The 2nd Workshop on Simulations for Biology and Chemistry Simbiochem 2026 SIMBIOCHEM II is a NeurIPS 2026 workshop on Machine Learning for Simulations in Biology & Chemistry that brings together ML, computational chemistry, biophysics and materials science to fuse simulation, generative models and agentic AI into physics-aligned systems that learn from reality. Learned potentials & force fields Differentiable & enhanced molecular dynamics Conformational ensembles, kinetics & rare events Molecular foundation models & scientific post-training Physical alignment: calibrated uncertainty & free energies Agentic ML: tool-calling MD/QM, active learning, closed-loop discovery Physically-grounded architectures (symmetries, conservation laws) Differentiable simulation & inverse design Data efficiency & scalability, foundation models Standardisation & reproducibility: benchmarks, datasets, metrics SIMBIOCHEM II — Machine Learning for Simulations in Biology & Chemistry\nThe 2nd SIMBIOCHEM Workshop\nNeurIPS 2026 Workshop · Sydney, Australia · Dec 11 or 12, 2026\nFusing molecular simulation, generative models and agentic AI into physics-aligned systems that learn from reality.\n\nThe workshop\nSimulation as the substrate for scientific AI\nThe next step beyond structure prediction is motion — trajectories, ensembles, kinetics and rare events. SIMBIOCHEM II brings together machine learning, computational chemistry, biophysics and materials science to make molecular AI faster, and grounded in physical rigour.\n\nThe guiding question: \"How can simulation, generative models, and agentic AI be fused into physics-aligned systems that learn from reality and accelerate biological and chemical discovery?\"\n\nThe workshop bridges machine learning with physical simulation, integrating scalable ML with rigorous physical simulation to make methods faster, and more accurate, reliable and grounded in science.\n\nTopics & themes\n- Learned potentials & force fields\n- Differentiable & enhanced molecular dynamics\n- Conformational ensembles, kinetics & rare events\n- Molecular foundation models & scientific post-training\n- Physical alignment: calibrated uncertainty & free energies\n- Agentic ML: tool-calling MD/QM, active learning, closed-loop discovery\n- Physically-grounded architectures (symmetries, conservation laws)\n- Differentiable simulation & inverse design\n- Next-generation learned potentials\n- Data efficiency & scalability, foundation models, agentic ML\n- Standardisation & reproducibility: benchmarks, datasets, metrics\n\nWho Can Submit\nA broad range of contributions across ML, computational chemistry, biophysics and materials science are welcome, particularly work showing how ML can accelerate physical simulation, and how simulation can inform and strengthen ML. Community: ML Researchers, Computational Chemistry, Biology, Materials Science, Biophysics, Life Sciences.\n\nReview Criteria\nSubmissions are evaluated on: Novelty (a new idea or genuinely new angle); Impact (does it matter to the field?); Correctness (are claims supported by evidence?); Clarity (can readers follow the method?).\n\nSubmission Format & Page Limits\nNon-archival short papers (5–8 pages) and abstracts, double-blind. Papers: 5–8 pages, excluding references, appendices and data-availability statements. References and appendices may be unlimited. Reviewers reserve the right to stop reading beyond 8 pages; extra content is at the author's risk. Six spotlight papers are selected for 10-minute talks; others present posters (24\" W × 36\" H, portrait orientation). Best-paper awards are given from accepted submissions.\n\nAnonymization Requirements\nSubmissions must be fully anonymised as per the template instructions. Do not include author names, affiliations or acknowledgements. Authors must remove any identifying links — GitHub repositories, personal websites, and images/figures (or their metadata) that could reveal your identity. Failing to submit an anonymised paper will cause desk rejection.\n\nConflict of Interest Policy\nThis year, NeurIPS requires disclosure of conflicts of interest. Authors must declare relationships with any organising or advisory committee member, including: same institution (current or past 3 years); PhD advisor/advisee relationships (any time); current or recent collaboration/co-authorship (past 3 years); family or close personal relationships; shared grants, funding, or acknowledgements; any other fairness-compromising relationship. Submissions from organisers, and from those with a personal COI to an organiser (their students, postdocs, close collaborators, family or close personal relationships), are not eligible.\n\nNon-archival policy\nThe workshop is non-archival, with no proceedings and no rebuttal phase. Camera-ready accepted papers are collated into a proceedings page on the website (authors may opt out). Authors may submit work previously presented at other workshops or conferences, or currently under review elsewhere.\n\nSubmission Platform\nSubmit via OpenReview at https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/Simbiochem\n\nKey Dates\n- Submission Deadline: August 29, 2026, 11:59 PM UTC\n- Author Notification: September 29, 2026 (AoE)\n- Workshop Day: Dec 11 or 12, 2026 (TBC by NeurIPS)\n- Camera-ready (accepted only): Date announced with decisions" }, { "key": "MusIML", "title": "The 7th Muslims in ML (MusIML) Workshop at NeurIPS 2026", "subtitle": "NeurIPS 2026 Workshop MusIML", "summary": "The 7th Muslims in ML (MusIML) Workshop, co-located with NeurIPS 2026 in Sydney, amplifies the voices of Muslim researchers in machine learning and AI, addressing both the potential for advancement and harm to Muslims and Muslim-majority communities, and fostering research, mentorship, and dialogue on ethical and inclusive AI.", "cfp_full": "Muslims in ML Community\nMuslims In ML (MusIML) is a community of Muslim researchers dedicated to amplifying the voices of Muslim researchers in the fields of machine learning and artificial intelligence and addressing challenges and research topics that are particularly relevant to Muslims. We focus on both the potential for advancement and harm to Muslims and those in Muslim-majority countries who religiously identify, culturally associate, or are classified by proximity, as \"Muslim\".\n\nOur next workshop, The 7th Muslims in ML Workshop, will be held in Sydney, Australia at NeurIPS 2026. The workshop has grown from approximately 100 attendees at our inaugural workshop at NeurIPS 2020 to accepting 60 papers with 300+ attendees at NeurIPS 2025.\n\nOur Vision\nOur vision is to create more equitable and impactful scientific contributions by Muslim communities in machine learning to foster meaningful research and innovation. We envision a world in which Muslims and those in Muslim-majority regions are empowered to engage in, contribute to, and benefit from global scientific discourse in machine learning. Through high-quality research, sustained dialogue, and cross-sector collaboration, we seek to advance research and promote researchers from Muslim backgrounds to meaningfully shape the development of AI systems that serve global society.\n\nOur Mission\nMuslims in ML provides a platform for the machine learning community to advance research, foster dialogue, and build collaborations at the intersection of machine learning, ethical AI, and Muslim communities and contexts. We bring together researchers, professionals, practitioners, and students to promote ML research carried out by members of the global Muslim community and conversations about AI's impact on Muslim communities worldwide. Grounded in advocacy, inclusivity, rigor, and integrity, we increase the visibility of trainees' work and support students by connecting them with mentors and collaborators across academia and industry.\n\nOur Core Values: Knowledge Seeking (Talab al-'Ilm); Excellence (Tafawuq); Innovation (Ibtikar); Collaboration (Mosharaka); Integrity (Nazaha); Inclusivity (Shumulia).\n\nCall for Papers — Five Submission Tracks\n- Track 1: ML for Muslim Communities — Research on Quranic analysis, Islamic digital humanities, language technologies, healthcare, and fairness (4-8 pages).\n- Track 2: AI System Demonstrations — Working prototypes addressing Islamic contexts, including a 2-page short or 4-page long demonstration paper plus video submission.\n- Track 3: ML Competition Proposals — Challenges requiring datasets with clearly defined evaluation metrics (max 2 pages).\n- Track 4: Mentored Research — Papers from mentorship programs or extended abstracts from major conferences (4-8 pages or max 2 pages).\n- Track 5: ML Research by Muslim Scholars — Any ML/AI topic with at least one Muslim author (4-8 pages).\n\nMusIML is non-archival by default, though accepted papers will appear on the workshop website and YouTube channel, with top papers invited for optional archival publication later. Workshop registration requires separate NeurIPS conference registration; complimentary conference registration may be available for select accepted paper authors based on factors including paper acceptance status, selection for oral or spotlight presentations, and career stage.\n\nImportant Dates\n- Call Opens: July 15, 2026\n- Visa-Friendly Deadline: August 10, 2026\n- Visa-Friendly Notification: August 20, 2026\n- Final Deadline: September 10, 2026\n- Final Notification: September 20, 2026\n- Camera-Ready: November 10, 2026", "cfp_status": "published", "topics": [ "ML for Muslim Communities (Quranic analysis, Islamic digital humanities, language technologies, healthcare, fairness)", "AI System Demonstrations addressing Islamic contexts", "ML Competition Proposals with datasets and defined evaluation metrics", "Mentored Research papers and extended abstracts", "ML Research by Muslim Scholars (any ML/AI topic with at least one Muslim author)", "Ethical AI and AI's impact on Muslim communities" ], "important_dates": [ { "label": "Call Opens", "date": "2026-07-15" }, { "label": "Visa-Friendly Deadline", "date": "2026-08-10" }, { "label": "Visa-Friendly Notification", "date": "2026-08-20" }, { "label": "Final Deadline", "date": "2026-09-10" }, { "label": "Final Notification", "date": "2026-09-20" }, { "label": "Camera-Ready", "date": "2026-11-10" } ], "organizers": [], "speakers": [], "host_url": "https://www.musiml.org/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/MusIML", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/MusIML", "location": "Sydney, Australia", "city": "Sydney", "workshop_date": "2026-12-05", "contact": "neurips@mail.musiml.org", "tracks": [ { "key": "MusIML", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/MusIML", "submission_dates_raw": "" } ], "group": "society", "group_label": "Society, Applications & Community", "corpus": "The 7th Muslims in ML (MusIML) Workshop at NeurIPS 2026 NeurIPS 2026 Workshop MusIML The 7th Muslims in ML (MusIML) Workshop, co-located with NeurIPS 2026 in Sydney, amplifies the voices of Muslim researchers in machine learning and AI, addressing both the potential for advancement and harm to Muslims and Muslim-majority communities, and fostering research, mentorship, and dialogue on ethical and inclusive AI. ML for Muslim Communities (Quranic analysis, Islamic digital humanities, language technologies, healthcare, fairness) AI System Demonstrations addressing Islamic contexts ML Competition Proposals with datasets and defined evaluation metrics Mentored Research papers and extended abstracts ML Research by Muslim Scholars (any ML/AI topic with at least one Muslim author) Ethical AI and AI's impact on Muslim communities Muslims in ML Community\nMuslims In ML (MusIML) is a community of Muslim researchers dedicated to amplifying the voices of Muslim researchers in the fields of machine learning and artificial intelligence and addressing challenges and research topics that are particularly relevant to Muslims. We focus on both the potential for advancement and harm to Muslims and those in Muslim-majority countries who religiously identify, culturally associate, or are classified by proximity, as \"Muslim\".\n\nOur next workshop, The 7th Muslims in ML Workshop, will be held in Sydney, Australia at NeurIPS 2026. The workshop has grown from approximately 100 attendees at our inaugural workshop at NeurIPS 2020 to accepting 60 papers with 300+ attendees at NeurIPS 2025.\n\nOur Vision\nOur vision is to create more equitable and impactful scientific contributions by Muslim communities in machine learning to foster meaningful research and innovation. We envision a world in which Muslims and those in Muslim-majority regions are empowered to engage in, contribute to, and benefit from global scientific discourse in machine learning. Through high-quality research, sustained dialogue, and cross-sector collaboration, we seek to advance research and promote researchers from Muslim backgrounds to meaningfully shape the development of AI systems that serve global society.\n\nOur Mission\nMuslims in ML provides a platform for the machine learning community to advance research, foster dialogue, and build collaborations at the intersection of machine learning, ethical AI, and Muslim communities and contexts. We bring together researchers, professionals, practitioners, and students to promote ML research carried out by members of the global Muslim community and conversations about AI's impact on Muslim communities worldwide. Grounded in advocacy, inclusivity, rigor, and integrity, we increase the visibility of trainees' work and support students by connecting them with mentors and collaborators across academia and industry.\n\nOur Core Values: Knowledge Seeking (Talab al-'Ilm); Excellence (Tafawuq); Innovation (Ibtikar); Collaboration (Mosharaka); Integrity (Nazaha); Inclusivity (Shumulia).\n\nCall for Papers — Five Submission Tracks\n- Track 1: ML for Muslim Communities — Research on Quranic analysis, Islamic digital humanities, language technologies, healthcare, and fairness (4-8 pages).\n- Track 2: AI System Demonstrations — Working prototypes addressing Islamic contexts, including a 2-page short or 4-page long demonstration paper plus video submission.\n- Track 3: ML Competition Proposals — Challenges requiring datasets with clearly defined evaluation metrics (max 2 pages).\n- Track 4: Mentored Research — Papers from mentorship programs or extended abstracts from major conferences (4-8 pages or max 2 pages).\n- Track 5: ML Research by Muslim Scholars — Any ML/AI topic with at least one Muslim author (4-8 pages).\n\nMusIML is non-archival by default, though accepted papers will appear on the workshop website and YouTube channel, with top papers invited for optional archival publication later. Workshop registration requires separate NeurIPS conference registration; complimentary conference registration may be available for select accepted paper authors based on factors including paper acceptance status, selection for oral or spotlight presentations, and career stage.\n\nImportant Dates\n- Call Opens: July 15, 2026\n- Visa-Friendly Deadline: August 10, 2026\n- Visa-Friendly Notification: August 20, 2026\n- Final Deadline: September 10, 2026\n- Final Notification: September 20, 2026\n- Camera-Ready: November 10, 2026" }, { "key": "MATH-AI", "title": "The Sixth Workshop on Mathematical Reasoning and AI", "subtitle": "NeurIPS 2026 Workshop MATH-AI", "summary": "MATH-AI is the sixth workshop on Mathematical Reasoning and AI, focusing this year on the intersection of agentic AI and mathematical reasoning - how AI systems can propose conjectures, formalize arguments, prove theorems, and collaborate reliably with human mathematicians across the mathematical research loop.", "cfp_full": "MATH-AI: The 6th Workshop on Mathematical Reasoning and AI\nNeurIPS 2026, Atlanta, December 12 or 13, 2026 (exact date & room TBA)\n\nOverview\nMathematical reasoning is central to science, engineering, finance, education, and mathematics itself. Since the first MATH-AI workshop, the field has moved from asking whether large language models (LLMs) can solve mathematical problems to asking how AI systems can participate across the full range of mathematical research: proposing conjectures, searching for examples and counterexamples, formalizing arguments, proving theorems, designing algorithms, and collaborating with human researchers.\n\nThis year, our workshop focuses more squarely on (but is not limited to) the intersection of agentic AI and mathematical reasoning. Recent progress makes this an especially timely moment: AI systems have achieved super-human results on competition-style and formal mathematical reasoning tasks, autoformalization is connecting natural mathematical language with proof assistants and formal libraries, and AI systems are beginning to guide mathematical discovery in topology, representation theory, combinatorics, matrix multiplication, and geometry. These advances point toward automated mathematical discovery, in which AI helps automate parts of the research loop from conjecture generation to proof search, verification, and communication.\n\nThis year, our central question is: How can agentic AI systems advance mathematical research while remaining reliable collaborators for human mathematicians? This theme links two priorities: building agents that can plan, use tools, conjecture, formalize, prove, verify, and learn from feedback; and designing human-AI workflows in which such agents extend mathematical judgment. It preserves the central question of previous MATH-AI workshops while reflecting the field's shift from isolated problem solving toward reliable mathematical agents and human-AI research workflows. To address this question, we aim to bring together diverse participants from different backgrounds, institutions, and disciplines into our workshop. Our objective is to foster a lively and constructive dialogue on areas related, but not limited, to the following:\n\n- Humans vs. machines: What are the comparative strengths, limitations, and characteristic failure modes of human mathematicians and AI systems? Which aspects of mathematical reasoning, judgment, and creativity remain distinctively human or machine?\n- Building reliable mathematical agents: What architectures, training methods, memory systems, tool interfaces, and feedback mechanisms enable agents to plan over long horizons, recover from errors, and operate reliably in mathematical environments?\n- Evaluating mathematical agents: How should we evaluate agents on advanced and open-ended mathematical tasks - beyond answer accuracy and formal correctness - to capture proof quality, robustness, creativity, efficiency, and effective use of tools and feedback?\n- Automated mathematical research: Which parts of the mathematical research process can agents conduct autonomously - from problem selection and experimentation to conjecture refinement and verification - and how should human researchers supervise, redirect, and collaborate with them?\n- Education: What roles can agentic AI systems play in mathematics education - tutoring, guiding exploration, and providing feedback - especially where expert instruction and educational resources are limited?\n- Cross-domain applications: How can mathematical agents enable progress in formal verification, software and hardware design, science, engineering, finance, and other domains that depend on complex mathematical reasoning?\n\nKey Dates\n- Paper submission opens: July 25, 2026\n- Paper submission deadline: September 25, 2026 (AoE)\n- Reviewing deadline: October 9, 2026 (AoE)\n- Author notification: October 19, 2026 (AoE)\n- Camera-ready deadline: October 29, 2026 (AoE)\n- Workshop: December 12 or 13, 2026 (TBA)\n\nThe call for papers is now open! Submit on OpenReview.\n\nContact: mathai.neurips2026@gmail.com", "cfp_status": "published", "topics": [ "Humans vs. machines: comparative strengths, limitations, and failure modes in mathematical reasoning", "Building reliable mathematical agents: architectures, training, memory, tool interfaces, and feedback", "Evaluating mathematical agents on advanced and open-ended tasks beyond answer accuracy", "Automated mathematical research: autonomous problem selection, experimentation, conjecture refinement, and verification", "Education: agentic AI for tutoring, guiding exploration, and feedback in mathematics", "Cross-domain applications: formal verification, software/hardware design, science, engineering, and finance" ], "important_dates": [ { "label": "Paper submission opens", "date": "2026-07-25" }, { "label": "Paper submission deadline (AoE)", "date": "2026-09-25" }, { "label": "Reviewing deadline (AoE)", "date": "2026-10-09" }, { "label": "Author notification (AoE)", "date": "2026-10-19" }, { "label": "Camera-ready deadline (AoE)", "date": "2026-10-29" }, { "label": "Workshop", "date": "2026-12-12" } ], "organizers": [ "Peiyang Song (CMU)", "Kaiyu Yang (Apodex)", "Pan Lu (Stanford)", "Patrick Shafto (DARPA & Rutgers)", "Sanjeev Arora (Princeton)", "Katie Collins (MIT & Princeton & Cambridge)", "Sean Welleck (CMU)", "Mateja Jamnik (Cambridge)" ], "speakers": [ "Yejin Choi (Stanford & NVIDIA)", "Sébastien Bubeck (OpenAI)", "Josh Tenenbaum (MIT)", "Timothy Gowers (Cambridge)", "Alex Gu (Math, Inc. & MIT)", "Graham Neubig (CMU & OpenHands)", "Jeremy Avigad (CMU)", "Kevin Ellis (Cornell)", "Nada Amin (Harvard)" ], "host_url": "https://mathai-2026.github.io/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/MATH-AI", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/MATH-AI", "location": "Atlanta, Georgia, United States", "city": "Atlanta", "workshop_date": "2026-12-12", "contact": "mathai.neurips2026@gmail.com", "tracks": [ { "key": "MATH-AI", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/MATH-AI", "submission_dates_raw": "" } ], "group": "theory", "group_label": "Theory & Methods", "corpus": "The Sixth Workshop on Mathematical Reasoning and AI NeurIPS 2026 Workshop MATH-AI MATH-AI is the sixth workshop on Mathematical Reasoning and AI, focusing this year on the intersection of agentic AI and mathematical reasoning - how AI systems can propose conjectures, formalize arguments, prove theorems, and collaborate reliably with human mathematicians across the mathematical research loop. Humans vs. machines: comparative strengths, limitations, and failure modes in mathematical reasoning Building reliable mathematical agents: architectures, training, memory, tool interfaces, and feedback Evaluating mathematical agents on advanced and open-ended tasks beyond answer accuracy Automated mathematical research: autonomous problem selection, experimentation, conjecture refinement, and verification Education: agentic AI for tutoring, guiding exploration, and feedback in mathematics Cross-domain applications: formal verification, software/hardware design, science, engineering, and finance MATH-AI: The 6th Workshop on Mathematical Reasoning and AI\nNeurIPS 2026, Atlanta, December 12 or 13, 2026 (exact date & room TBA)\n\nOverview\nMathematical reasoning is central to science, engineering, finance, education, and mathematics itself. Since the first MATH-AI workshop, the field has moved from asking whether large language models (LLMs) can solve mathematical problems to asking how AI systems can participate across the full range of mathematical research: proposing conjectures, searching for examples and counterexamples, formalizing arguments, proving theorems, designing algorithms, and collaborating with human researchers.\n\nThis year, our workshop focuses more squarely on (but is not limited to) the intersection of agentic AI and mathematical reasoning. Recent progress makes this an especially timely moment: AI systems have achieved super-human results on competition-style and formal mathematical reasoning tasks, autoformalization is connecting natural mathematical language with proof assistants and formal libraries, and AI systems are beginning to guide mathematical discovery in topology, representation theory, combinatorics, matrix multiplication, and geometry. These advances point toward automated mathematical discovery, in which AI helps automate parts of the research loop from conjecture generation to proof search, verification, and communication.\n\nThis year, our central question is: How can agentic AI systems advance mathematical research while remaining reliable collaborators for human mathematicians? This theme links two priorities: building agents that can plan, use tools, conjecture, formalize, prove, verify, and learn from feedback; and designing human-AI workflows in which such agents extend mathematical judgment. It preserves the central question of previous MATH-AI workshops while reflecting the field's shift from isolated problem solving toward reliable mathematical agents and human-AI research workflows. To address this question, we aim to bring together diverse participants from different backgrounds, institutions, and disciplines into our workshop. Our objective is to foster a lively and constructive dialogue on areas related, but not limited, to the following:\n\n- Humans vs. machines: What are the comparative strengths, limitations, and characteristic failure modes of human mathematicians and AI systems? Which aspects of mathematical reasoning, judgment, and creativity remain distinctively human or machine?\n- Building reliable mathematical agents: What architectures, training methods, memory systems, tool interfaces, and feedback mechanisms enable agents to plan over long horizons, recover from errors, and operate reliably in mathematical environments?\n- Evaluating mathematical agents: How should we evaluate agents on advanced and open-ended mathematical tasks - beyond answer accuracy and formal correctness - to capture proof quality, robustness, creativity, efficiency, and effective use of tools and feedback?\n- Automated mathematical research: Which parts of the mathematical research process can agents conduct autonomously - from problem selection and experimentation to conjecture refinement and verification - and how should human researchers supervise, redirect, and collaborate with them?\n- Education: What roles can agentic AI systems play in mathematics education - tutoring, guiding exploration, and providing feedback - especially where expert instruction and educational resources are limited?\n- Cross-domain applications: How can mathematical agents enable progress in formal verification, software and hardware design, science, engineering, finance, and other domains that depend on complex mathematical reasoning?\n\nKey Dates\n- Paper submission opens: July 25, 2026\n- Paper submission deadline: September 25, 2026 (AoE)\n- Reviewing deadline: October 9, 2026 (AoE)\n- Author notification: October 19, 2026 (AoE)\n- Camera-ready deadline: October 29, 2026 (AoE)\n- Workshop: December 12 or 13, 2026 (TBA)\n\nThe call for papers is now open! Submit on OpenReview.\n\nContact: mathai.neurips2026@gmail.com" }, { "key": "GenAI4Health", "title": "The Third Workshop on GenAI for Health: Agentic Systems, Clinical Trust, and Future Potential", "subtitle": "NeurIPS 2026 Workshop GenAI4Health", "summary": "GenAI4Health convenes machine learning researchers, healthcare professionals, policy experts, and clinical practitioners to advance capable, safe, equitable, and trusted generative AI for health, with a 2026 focus on agentic systems, clinical trust, and next-generation ambient, conversational, and embodied human-AI interaction for care.", "cfp_full": "About\nBuilding on successful prior editions, this third GenAI4Health workshop convenes machine learning researchers, healthcare professionals, policy experts, and clinical practitioners to advance capable, safe, equitable, and trusted generative AI for health. The 2026 edition introduces a forward-looking track on next-generation human-AI interaction, spanning ambient, conversational, and embodied systems for care. The workshop runs December 11-12, 2026, in Sydney, Australia.\n\nTopics of Interest\n\nTrack 1: Frontier Models for Health\nSubmissions should advance foundation model capabilities in healthcare contexts, including agentic systems, clinical reasoning, and multimodal learning. The workshop welcomes work on retrieval-augmented generation, diagnosis/prognosis systems, tool use in healthcare, clinical language models, and longitudinal patient modeling.\n\nTrack 2: Trustworthy AI, Policy, Human-AI Collaboration & Adoption\nThis track focuses on safety, robustness, and fairness of generative AI in medical contexts, and strategies for responsible real-world deployment. Topics include AI nurse copilots, surgical AI, safety benchmarks, red teaming, explainability, bias mitigation, human-AI collaboration, regulatory alignment, and real-world performance evaluation.\n\nTrack 3: Toward 360-degree AI Care - Ambient, Conversational & Embodied Systems\nA forward-looking track exploring systems that listen, converse, remember, and act across ambient intelligence, voice-first interfaces, and embodied agents. Topics include voice agents, ambient scribes, conversational AI for care delivery, virtual care navigators, multimodal interactive systems, and accessibility for aging populations.\n\nThree Submission Tracks\n\nResearch Papers\n- Page limit: Up to 9 pages (excluding acknowledgments, references, appendix)\n- The core of the program: methodological advances and empirical studies in generative AI for health.\n- Requirements: Complete, self-contained works with clearly stated problems, described methods, and empirical evidence supporting claims. Experiments should include appropriate baselines and evaluation protocols.\n- Evaluation criteria: Technical soundness, novelty, health relevance/impact, and presentation clarity.\n\nDemonstration Papers\n- Page limit: Up to 5 pages (excluding acknowledgments, references, appendix)\n- Working systems, applications, and tools relevant to generative AI in health.\n- Requirements: Functioning systems (deployed, piloted, or working prototype) with architecture, intended users, and use context described. A comprehensive quantitative evaluation is not required; however, submissions should include evidence that the system works as described.\n- Evaluation criteria: System functionality, relevance to health stakeholders, technical interest, and demonstration clarity.\n\nPosition Papers\n- Page limit: Up to 5 pages (excluding acknowledgments, references, appendix)\n- Perspectives, analyses, and proposals on policy, governance, evaluation practices, or deployment strategies for generative AI in health.\n- Requirements: Position papers should advance a clear, well-argued thesis. New experiments are not required, but arguments must be grounded in evidence.\n- Evaluation criteria: Position clarity and significance, argument quality/grounding, community discussion potential, and multidisciplinary relevance.\n\nSubmission Requirements\nSubmissions must use the NeurIPS 2026 LaTeX style file with the command \\usepackage{neurips_2026} to ensure anonymity. Papers that fail to meet submission requirements may be rejected without consideration of their merits.\nKey policies:\n- Papers submitted to the workshop must not have been previously published at another venue at the time of submission.\n- Supplementary materials (code, data, videos) may accompany submissions.\n- Accepted papers will be non-archival (NOT included in proceedings or any form of publication).\n- Authors may opt out of public availability on OpenReview before the camera-ready stage.\n\nReview Process\n- Platform: OpenReview (portal link to be posted)\n- Review approach: Double-blind review with anonymization strictly enforced\n- Number of reviews: Each submission receives at least two reviews\n- Author response: There is no author rebuttal or response stage\n- Confidentiality: Submissions visible only to reviewers and organizers; rejected submissions remain private\n\nPublication & Dual Submission Policy\nGenAI4Health is non-archival, meaning accepted papers will not be published in any proceedings and will not be considered a formal (archival) publication.\n- You may submit your workshop paper to other venues afterwards.\n- Work under review elsewhere may be submitted here (if the other venue permits).\n- Previously published work in archival venues is ineligible; non-archival preprints are acceptable.\n- Accepted papers will be made publicly available on OpenReview by default after the camera-ready stage.\n\nAuthorship Policy\n- No authors may be added after the submission deadline.\n- Author reordering is permitted for accepted papers with consent.\n- Author removal requires written consent from all authors and must be submitted to program chairs.\n\nPresentation & Awards\nAll accepted papers will be presented as posters. Oral/spotlight presentations will be selected from the accepted papers. Three Outstanding Paper Awards will be selected, one per track.\n\nImportant Dates\n- Paper Submission Deadline: September 5, 2026\n- Acceptance Notification: September 29, 2026\n- Camera-ready Submission: TBA\n- Workshop Day: TBA (December 11-12, 2026, Sydney)\nAll deadlines are at 11:59 PM AoE (Anywhere on Earth).", "cfp_status": "published", "topics": [ "Frontier/foundation models for health: agentic systems, clinical reasoning, and multimodal learning", "Retrieval-augmented generation for healthcare", "Diagnosis/prognosis systems", "Tool use in healthcare", "Clinical language models", "Longitudinal patient modeling", "Safety, robustness, and fairness of generative AI in medical contexts", "AI nurse copilots", "Surgical AI", "Safety benchmarks and red teaming", "Explainability and bias mitigation", "Human-AI collaboration", "Regulatory alignment and real-world performance evaluation", "Ambient intelligence, voice-first interfaces, and embodied agents for care", "Voice agents and ambient scribes", "Conversational AI for care delivery", "Virtual care navigators", "Multimodal interactive systems", "Accessibility for aging populations" ], "important_dates": [ { "label": "Paper Submission Deadline (all tracks, AoE)", "date": "2026-09-05" }, { "label": "Acceptance Notification", "date": "2026-09-29" }, { "label": "Camera-ready Submission", "date": "TBA" }, { "label": "Workshop", "date": "2026-12-11" } ], "organizers": [ "Jiayuan Ding (Hippocratic AI)", "Pranav Rajpurkar (Harvard University)", "Junyuan Hong (National University of Singapore)", "Joseph J. Lim (KAIST)", "Tanveer Fathima Syeda-Mahmood (Stanford University)", "Ehsan Adeli (Stanford University)", "Subhabrata Mukherjee (Hippocratic AI)", "Fei-Fei Li (Stanford University)", "Ying Ding (UT Austin)", "Jiawei Xu (UT Austin, Student Organizer)", "Jinrui Fang (UT Austin, Student Organizer)", "Ziheng Zhang (UT Austin, Student Organizer)", "Tiange Xiang (Stanford University, Student Organizer)", "Yixin Wang (Stanford University, Student Organizer)", "Hyewon Jeong (MIT, Student Organizer)", "Sheng Liu (Stanford University, Local Organizing Committee)", "Kevin Y.C. Hou (Sydney Medical School, Local Organizing Committee)", "Matthew Shu (University of Sydney, Local Organizing Committee)" ], "speakers": [ "Pushmeet Kohli (Google DeepMind)", "Karan Singhal (OpenAI)", "James Zou (Stanford University)", "Jinman Kim (University of Sydney)", "Haider Warraich (ARPA-H)", "Danielle Bitterman (Harvard Medical School)", "Enrico Coiera (Australian Institute of Health Innovation)", "Liliana Laranjo (University of Sydney)", "Ming-Yu Liu (NVIDIA)", "Sergey Levine (UC Berkeley)", "Chelsea Finn (Stanford University)", "Hoifung Poon (Recursion)" ], "host_url": "https://genai4health.github.io/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/GenAI4Health", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/GenAI4Health", "location": "Sydney, Australia", "city": "Sydney", "workshop_date": "2026-12-11", "contact": "genai4health@googlegroups.com", "tracks": [ { "key": "GenAI4Health", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/GenAI4Health", "submission_dates_raw": "" }, { "key": "GenAI4Health_Demonstration_Paper_Track", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/GenAI4Health_Demonstration_Paper_Track", "submission_dates_raw": "" }, { "key": "GenAI4Health_Position_Paper_Track", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/GenAI4Health_Position_Paper_Track", "submission_dates_raw": "" } ], "group": "health", "group_label": "Health & Medicine", "corpus": "The Third Workshop on GenAI for Health: Agentic Systems, Clinical Trust, and Future Potential NeurIPS 2026 Workshop GenAI4Health GenAI4Health convenes machine learning researchers, healthcare professionals, policy experts, and clinical practitioners to advance capable, safe, equitable, and trusted generative AI for health, with a 2026 focus on agentic systems, clinical trust, and next-generation ambient, conversational, and embodied human-AI interaction for care. Frontier/foundation models for health: agentic systems, clinical reasoning, and multimodal learning Retrieval-augmented generation for healthcare Diagnosis/prognosis systems Tool use in healthcare Clinical language models Longitudinal patient modeling Safety, robustness, and fairness of generative AI in medical contexts AI nurse copilots Surgical AI Safety benchmarks and red teaming Explainability and bias mitigation Human-AI collaboration Regulatory alignment and real-world performance evaluation Ambient intelligence, voice-first interfaces, and embodied agents for care Voice agents and ambient scribes Conversational AI for care delivery Virtual care navigators Multimodal interactive systems Accessibility for aging populations About\nBuilding on successful prior editions, this third GenAI4Health workshop convenes machine learning researchers, healthcare professionals, policy experts, and clinical practitioners to advance capable, safe, equitable, and trusted generative AI for health. The 2026 edition introduces a forward-looking track on next-generation human-AI interaction, spanning ambient, conversational, and embodied systems for care. The workshop runs December 11-12, 2026, in Sydney, Australia.\n\nTopics of Interest\n\nTrack 1: Frontier Models for Health\nSubmissions should advance foundation model capabilities in healthcare contexts, including agentic systems, clinical reasoning, and multimodal learning. The workshop welcomes work on retrieval-augmented generation, diagnosis/prognosis systems, tool use in healthcare, clinical language models, and longitudinal patient modeling.\n\nTrack 2: Trustworthy AI, Policy, Human-AI Collaboration & Adoption\nThis track focuses on safety, robustness, and fairness of generative AI in medical contexts, and strategies for responsible real-world deployment. Topics include AI nurse copilots, surgical AI, safety benchmarks, red teaming, explainability, bias mitigation, human-AI collaboration, regulatory alignment, and real-world performance evaluation.\n\nTrack 3: Toward 360-degree AI Care - Ambient, Conversational & Embodied Systems\nA forward-looking track exploring systems that listen, converse, remember, and act across ambient intelligence, voice-first interfaces, and embodied agents. Topics include voice agents, ambient scribes, conversational AI for care delivery, virtual care navigators, multimodal interactive systems, and accessibility for aging populations.\n\nThree Submission Tracks\n\nResearch Papers\n- Page limit: Up to 9 pages (excluding acknowledgments, references, appendix)\n- The core of the program: methodological advances and empirical studies in generative AI for health.\n- Requirements: Complete, self-contained works with clearly stated problems, described methods, and empirical evidence supporting claims. Experiments should include appropriate baselines and evaluation protocols.\n- Evaluation criteria: Technical soundness, novelty, health relevance/impact, and presentation clarity.\n\nDemonstration Papers\n- Page limit: Up to 5 pages (excluding acknowledgments, references, appendix)\n- Working systems, applications, and tools relevant to generative AI in health.\n- Requirements: Functioning systems (deployed, piloted, or working prototype) with architecture, intended users, and use context described. A comprehensive quantitative evaluation is not required; however, submissions should include evidence that the system works as described.\n- Evaluation criteria: System functionality, relevance to health stakeholders, technical interest, and demonstration clarity.\n\nPosition Papers\n- Page limit: Up to 5 pages (excluding acknowledgments, references, appendix)\n- Perspectives, analyses, and proposals on policy, governance, evaluation practices, or deployment strategies for generative AI in health.\n- Requirements: Position papers should advance a clear, well-argued thesis. New experiments are not required, but arguments must be grounded in evidence.\n- Evaluation criteria: Position clarity and significance, argument quality/grounding, community discussion potential, and multidisciplinary relevance.\n\nSubmission Requirements\nSubmissions must use the NeurIPS 2026 LaTeX style file with the command \\usepackage{neurips_2026} to ensure anonymity. Papers that fail to meet submission requirements may be rejected without consideration of their merits.\nKey policies:\n- Papers submitted to the workshop must not have been previously published at another venue at the time of submission.\n- Supplementary materials (code, data, videos) may accompany submissions.\n- Accepted papers will be non-archival (NOT included in proceedings or any form of publication).\n- Authors may opt out of public availability on OpenReview before the camera-ready stage.\n\nReview Process\n- Platform: OpenReview (portal link to be posted)\n- Review approach: Double-blind review with anonymization strictly enforced\n- Number of reviews: Each submission receives at least two reviews\n- Author response: There is no author rebuttal or response stage\n- Confidentiality: Submissions visible only to reviewers and organizers; rejected submissions remain private\n\nPublication & Dual Submission Policy\nGenAI4Health is non-archival, meaning accepted papers will not be published in any proceedings and will not be considered a formal (archival) publication.\n- You may submit your workshop paper to other venues afterwards.\n- Work under review elsewhere may be submitted here (if the other venue permits).\n- Previously published work in archival venues is ineligible; non-archival preprints are acceptable.\n- Accepted papers will be made publicly available on OpenReview by default after the camera-ready stage.\n\nAuthorship Policy\n- No authors may be added after the submission deadline.\n- Author reordering is permitted for accepted papers with consent.\n- Author removal requires written consent from all authors and must be submitted to program chairs.\n\nPresentation & Awards\nAll accepted papers will be presented as posters. Oral/spotlight presentations will be selected from the accepted papers. Three Outstanding Paper Awards will be selected, one per track.\n\nImportant Dates\n- Paper Submission Deadline: September 5, 2026\n- Acceptance Notification: September 29, 2026\n- Camera-ready Submission: TBA\n- Workshop Day: TBA (December 11-12, 2026, Sydney)\nAll deadlines are at 11:59 PM AoE (Anywhere on Earth)." }, { "key": "ATTRIB", "title": "Third NeurIPS Workshop on Attributing Model Behavior at Scale: Data Attribution and Provenance", "subtitle": "ATTRIB 2026", "summary": "ATTRIB brings together the contributive and corroborative data-attribution communities with practitioners in law, journalism, music, and AI safety to make attribution of model outputs interpretable, auditable, and grounded in identifiable sources.", "cfp_full": "How can we attribute model outputs and actions back to data? How can we design attributions that inform downstream uses like alignment, data use litigation, and regulatory audits?\n\nAs generative AI permeates rapidly across society, an increasingly relevant problem is attribution: how can we attribute model outputs and actions back to data? The appropriate notion of attribution depends on the specific applications. Contributive attribution aims to determine which training data causally influenced the generated output. A growing body of work has made this increasingly tractable, though challenges remain. In parallel, corroborative attribution methods leverage lexical techniques, conditional generation, and textual entailment to identify sources that textually entail, or otherwise semantically correspond to, a model output, i.e., citations.\n\nHowever, contributive methods produce scores that are often difficult to interpret or verify and corroborative methods identify semantically supporting sources but make no claims about causal responsibility. In practice, however, stakeholders in law, journalism, and the arts need attribution that is interpretable, auditable, and grounded in identifiable sources, beyond computability. Meanwhile, the rapid growth of synthetic and model-generated data further complicates attribution by blurring the boundary between training data and model outputs. Bridging the gap between methods and applications will require not only technical innovation but also clearer problem formulations informed by the concrete constraints of real applications.\n\nThis workshop will bring together researchers from both the contributive and corroborative attribution communities with practitioners in law, music, journalism, and AI safety. Our goal is to identify where existing methods fall short of practical needs, surface shared technical challenges across application domains, and establish concrete research directions that can move attribution from a purely academic tool toward one that is useful in practice.\n\nCall for Papers\n\nSubmissions open August 1st!\n\nWe are soliciting papers along two tracks:\n- Main track papers: 3-6 page submissions on attributing model behaviors (see below for example topics).\n- Idea track papers: 2-4 page submissions, on a specific topic or on the field of attribution as a whole. Vision papers, unifying ideas, documentation of failed experiments are all welcome. Papers in this track will be held to the same standard as the main track, but can be opinionated (so long as opinions are clearly demarcated from facts) and need not have experiments (although lack of experimentation should be justified).\n\nAlong these tracks, we welcome submissions pertaining to any aspect of model behavior attribution. For example:\n- Contributive attribution: Contributive attribution aims to estimate the causal effect of training data on model behavior. While methods exist, key questions remain: How reliable and robust are current approaches, and how can their outputs be verified or audited in practice?\n- Corroborative attribution: Corroborative attribution identifies training data that supports a given generation. What makes a good corroborative attribution method, and how do we evaluate them? How can corroborative attribution be verified and made meaningful for downstream uses?\n- Downstream applications of data attribution: Data attribution has potential beyond technical uses like data selection. What other domains can it inform? Can attribution methods contribute to broader goals of accountability and trustworthy AI, and if so, what form must they take to be useful in those contexts?\n- Attribution of synthetic data: The growing use of synthetic data in model training complicates both legal and technical notions of attribution. When training data is itself model-generated, what are the implications for copyright, provenance, and liability? How can model behavior be traced back to the original data that shaped it?\n\nSubmission Instructions\n\nFormat submissions as follows:\n- 3-6 pages (main track) or 2-4 pages (idea track)\n- NeurIPS 2026 paper formatting (download from the NeurIPS 2026 Call for Papers)\n- Appendix included in the same PDF as the main body\n- No Appendix page limit\n\nATTRIB uses a reciprocal reviewing model: for each submission, at least one author is expected to serve as a reviewer. Reviewers are assigned up to 2 papers, with reviews due September 22 (AOE). Please designate your submission's reciprocal reviewer on the Reviewer Registration Form linked in the OpenReview portal. We rely on authors to make the review process work, and submissions without a participating reviewer may be desk rejected.\n\nWhen ready, submit to OpenReview (the workshop is non-archival).\n\nLLM Usage Policy\n\nATTRIB uses the same LLM usage policy as NeurIPS 2026. Please see the NeurIPS 2026 Main Track Handbook for details.\n\nImportant Dates\n- August 1: Submission portal opens\n- September 1 (AOE): Deadline for both idea and main track papers\n- September 22 (AOE): Deadline for Reviews\n- September 29: Decision notifications", "cfp_status": "published", "topics": [ "Contributive attribution (estimating the causal effect of training data on model behavior; reliability, robustness, verification, auditing)", "Corroborative attribution (identifying training data that supports a generation; evaluation and verification of citations)", "Downstream applications of data attribution (accountability, trustworthy AI, beyond data selection)", "Attribution of synthetic data (copyright, provenance, and liability when training data is model-generated)" ], "important_dates": [ { "label": "Submission portal opens", "date": "2026-08-01" }, { "label": "Submission Deadline (idea and main track)", "date": "2026-09-01" }, { "label": "Deadline for Reviews", "date": "2026-09-22" }, { "label": "Decision notifications", "date": "2026-09-29" } ], "organizers": [ "Bálint Mucsányi", "Sarah H. Cen", "Andrew Ilyas", "Elisa Nguyen", "Sung Min Park", "Teddi Worledge" ], "speakers": [ "A. Feder Cooper (Incoming Assistant Professor at Yale Computer Science)", "Jiaqi Ma (Assistant Professor at UIUC Information Sciences)", "Sewon Min (Assistant Professor at UC Berkeley EECS)", "Aileen Nielsen (Visiting Assistant Professor at Harvard Law School)", "John Thickstun (Assistant Professor at Cornell Computer Science)" ], "host_url": "https://attrib-workshop.cc/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/ATTRIB", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/ATTRIB", "location": "Sydney, Australia", "city": "Sydney", "workshop_date": "2026-12-10", "contact": "attrib-neurips26@googlegroups.com", "tracks": [ { "key": "ATTRIB", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/ATTRIB", "submission_dates_raw": "" } ], "group": "trust", "group_label": "Trustworthy & Safe AI", "corpus": "Third NeurIPS Workshop on Attributing Model Behavior at Scale: Data Attribution and Provenance ATTRIB 2026 ATTRIB brings together the contributive and corroborative data-attribution communities with practitioners in law, journalism, music, and AI safety to make attribution of model outputs interpretable, auditable, and grounded in identifiable sources. Contributive attribution (estimating the causal effect of training data on model behavior; reliability, robustness, verification, auditing) Corroborative attribution (identifying training data that supports a generation; evaluation and verification of citations) Downstream applications of data attribution (accountability, trustworthy AI, beyond data selection) Attribution of synthetic data (copyright, provenance, and liability when training data is model-generated) How can we attribute model outputs and actions back to data? How can we design attributions that inform downstream uses like alignment, data use litigation, and regulatory audits?\n\nAs generative AI permeates rapidly across society, an increasingly relevant problem is attribution: how can we attribute model outputs and actions back to data? The appropriate notion of attribution depends on the specific applications. Contributive attribution aims to determine which training data causally influenced the generated output. A growing body of work has made this increasingly tractable, though challenges remain. In parallel, corroborative attribution methods leverage lexical techniques, conditional generation, and textual entailment to identify sources that textually entail, or otherwise semantically correspond to, a model output, i.e., citations.\n\nHowever, contributive methods produce scores that are often difficult to interpret or verify and corroborative methods identify semantically supporting sources but make no claims about causal responsibility. In practice, however, stakeholders in law, journalism, and the arts need attribution that is interpretable, auditable, and grounded in identifiable sources, beyond computability. Meanwhile, the rapid growth of synthetic and model-generated data further complicates attribution by blurring the boundary between training data and model outputs. Bridging the gap between methods and applications will require not only technical innovation but also clearer problem formulations informed by the concrete constraints of real applications.\n\nThis workshop will bring together researchers from both the contributive and corroborative attribution communities with practitioners in law, music, journalism, and AI safety. Our goal is to identify where existing methods fall short of practical needs, surface shared technical challenges across application domains, and establish concrete research directions that can move attribution from a purely academic tool toward one that is useful in practice.\n\nCall for Papers\n\nSubmissions open August 1st!\n\nWe are soliciting papers along two tracks:\n- Main track papers: 3-6 page submissions on attributing model behaviors (see below for example topics).\n- Idea track papers: 2-4 page submissions, on a specific topic or on the field of attribution as a whole. Vision papers, unifying ideas, documentation of failed experiments are all welcome. Papers in this track will be held to the same standard as the main track, but can be opinionated (so long as opinions are clearly demarcated from facts) and need not have experiments (although lack of experimentation should be justified).\n\nAlong these tracks, we welcome submissions pertaining to any aspect of model behavior attribution. For example:\n- Contributive attribution: Contributive attribution aims to estimate the causal effect of training data on model behavior. While methods exist, key questions remain: How reliable and robust are current approaches, and how can their outputs be verified or audited in practice?\n- Corroborative attribution: Corroborative attribution identifies training data that supports a given generation. What makes a good corroborative attribution method, and how do we evaluate them? How can corroborative attribution be verified and made meaningful for downstream uses?\n- Downstream applications of data attribution: Data attribution has potential beyond technical uses like data selection. What other domains can it inform? Can attribution methods contribute to broader goals of accountability and trustworthy AI, and if so, what form must they take to be useful in those contexts?\n- Attribution of synthetic data: The growing use of synthetic data in model training complicates both legal and technical notions of attribution. When training data is itself model-generated, what are the implications for copyright, provenance, and liability? How can model behavior be traced back to the original data that shaped it?\n\nSubmission Instructions\n\nFormat submissions as follows:\n- 3-6 pages (main track) or 2-4 pages (idea track)\n- NeurIPS 2026 paper formatting (download from the NeurIPS 2026 Call for Papers)\n- Appendix included in the same PDF as the main body\n- No Appendix page limit\n\nATTRIB uses a reciprocal reviewing model: for each submission, at least one author is expected to serve as a reviewer. Reviewers are assigned up to 2 papers, with reviews due September 22 (AOE). Please designate your submission's reciprocal reviewer on the Reviewer Registration Form linked in the OpenReview portal. We rely on authors to make the review process work, and submissions without a participating reviewer may be desk rejected.\n\nWhen ready, submit to OpenReview (the workshop is non-archival).\n\nLLM Usage Policy\n\nATTRIB uses the same LLM usage policy as NeurIPS 2026. Please see the NeurIPS 2026 Main Track Handbook for details.\n\nImportant Dates\n- August 1: Submission portal opens\n- September 1 (AOE): Deadline for both idea and main track papers\n- September 22 (AOE): Deadline for Reviews\n- September 29: Decision notifications" }, { "key": "AIWILD", "title": "Third Workshop on Agents in the Wild: Safety, Security, and Beyond", "subtitle": "NeurIPS 2026 AIWILD", "summary": "The third edition of the Agents in the Wild workshop (following ICLR 2026 and ICML 2026), bringing together academia and industry on methods, benchmarks, and frameworks for building reliable and trustworthy AI agents that can reason, act, and adapt safely and securely in open-ended real-world environments.", "cfp_full": "About\nAI agents are being rapidly deployed in the real world — from Claude Code and OpenAI Codex to open-source tools like OpenClaw — yet safety & security research still lags behind. As these systems reason, act, and adapt in open-ended environments, they introduce profound and emerging challenges around safety, security, and trustworthiness that the research community is only beginning to face.\n\nThe urgency is only intensifying. In July 2026, OpenAI models escaped a sandboxed evaluation and compromised Hugging Face's production systems on their own. Governments are engaging as well: in June 2026 the U.S. government imposed — then weeks later lifted — export controls restricting access to Anthropic's most capable models over national-security concerns. At the same time, the debate over what open- and closed-weight models can safely do has sharpened, with NVIDIA CEO Jensen Huang's open-weights letter arguing that open models strengthen safety and security, and more than a thousand employees of frontier AI companies calling for tools to deliberately pace automated AI development.\n\nThis workshop brings together researchers and practitioners across academia and industry to chart the next steps toward reliable agents that can operate responsibly in the wild. We build on the momentum of our first workshop at ICLR 2026 (237 submissions, 300+ attendees) and our second workshop at ICML 2026 (327 submissions, 300+ attendees), and this third edition sharpens the agenda around emerging emphases.\n\nCall for Papers\nThe Third Workshop on Agents in the Wild at NeurIPS 2026 invites submissions from researchers and practitioners exploring how intelligent agents can reason, act, and adapt safely and securely in open-ended real-world environments.\n\nAs agentic AI systems grow more capable, their deployment in dynamic, unpredictable settings introduces challenges in safety, security, and general trustworthiness. This workshop aims to spark discussion across academia and industry on methods, benchmarks, and frameworks for building reliable and trustworthy agents that can operate responsibly \"in the wild.\"\n\nScope\nWe welcome contributions on a wide range of topics related to AI agents, including but not limited to:\n- Agent alignment, control, and oversight\n- Agentic attack surface and defenses\n- Privacy, robustness, and factuality for agents\n- Agentic interpretability and fairness\n- Evaluation and benchmarking agents\n- Multimodal and computer-use agent safety\n- Multi-agent coordination and long-horizon reliability\n- Post-training and adaptation of agents\n- Agent infrastructure and protocol security\n- Agent skills and applications\n- Agent identity and accountability\n- Safety and security of recursive self-improvement and autonomous research agents\n- Long-horizon, persistent, and continually-learning agents\n- Tool, MCP, and CLI/coding-agent security\n- AI agents for science and autonomous discovery\n- Human–agent collaboration and co-work across apps and tasks\n- Broader societal governance considerations\n\nImportant Dates\nAll dates are tentative and subject to change.\n- Paper Submission Open: August 1, 2026 AoE\n- Paper Submission Deadline: August 29, 2026 AoE\n- Paper Notification Deadline: September 29, 2026 AoE\n- Camera-ready Version Deadline: December 4, 2026 AoE\n- Workshop Date: December 11 or 12, 2026 (Sydney, Australia)\n\nSubmission Guidelines\nFormat: The workshop will have a single track. Submissions follow the NeurIPS 2026 template (up to 9 pages). References and supplementary materials will not count against the page limit. Each submission will receive at least 3 reviews.\n\nSubmission site: Submit papers through the Workshop Submission Portal on OpenReview.\n\nOpenReview profiles: If you do not already have an OpenReview account, please create your profile at least two weeks in advance of the paper submission deadline (i.e., by August 15, 2026), as new profiles can take time to be activated.\n\nStyle file: Please format your submission using the NeurIPS 2026 LaTeX style file. A modified template that refers to our workshop will be made available here. Submissions that exceed the page limit may be desk-rejected.\n\nDual-submission policy: The workshop will adopt a non-archival policy, welcoming ongoing and unpublished work, as well as papers under review or recently accepted at other venues (provided they do not breach dual-submission or anonymity policies of the other venue). We discourage the submission of work previously published at major venues (e.g., NeurIPS, ICLR, ICML).\n\nVisibility: Accepted papers will be made public, but rejected submissions and reviews will not.\n\nDouble-blind reviewing: Submissions must be fully anonymized. This policy applies to any supplementary or linked material as well, including code. Any papers found to be in violation of this policy may be desk-rejected.\n\nConflicts of interest: The organizing committee will proactively search for conflicts of interest using the OpenReview profiles of authors and reviewers, ensuring that reviewers are not assigned any submissions from their own organization. Members of the organizing committee will not be involved in the assessment of any submission from the same organization. We will also not accept submissions from workshop organizers or any person having a personal conflict of interest with them.\n\nLLM usage policy: For LLM usage, authors should follow the NeurIPS 2026 Main Track Handbook.\n\nContact: For any questions, please contact us at agents-in-the-wild-neurips2026@googlegroups.com.", "cfp_status": "published", "topics": [ "Agent alignment, control, and oversight", "Agentic attack surface and defenses", "Privacy, robustness, and factuality for agents", "Agentic interpretability and fairness", "Evaluation and benchmarking agents", "Multimodal and computer-use agent safety", "Multi-agent coordination and long-horizon reliability", "Post-training and adaptation of agents", "Agent infrastructure and protocol security", "Agent skills and applications", "Agent identity and accountability", "Safety and security of recursive self-improvement and autonomous research agents", "Long-horizon, persistent, and continually-learning agents", "Tool, MCP, and CLI/coding-agent security", "AI agents for science and autonomous discovery", "Human–agent collaboration and co-work across apps and tasks", "Broader societal governance considerations" ], "important_dates": [ { "label": "Paper Submission Open", "date": "2026-08-01" }, { "label": "Paper Submission Deadline", "date": "2026-08-29" }, { "label": "Paper Notification Deadline", "date": "2026-09-29" }, { "label": "Camera-ready Version Deadline", "date": "2026-12-04" }, { "label": "Workshop Date", "date": "2026-12-11 or 2026-12-12" } ], "organizers": [ "Chenguang Wang (UC Santa Cruz)", "Wei Wang (UCLA)", "Yuan (Emily) Xue (Scale AI)", "Nicholas Crispino (UC Santa Cruz)", "Tianneng Shi (UC Berkeley)", "Vincent Siu (UC Santa Cruz)", "Zhe Ye (UC Berkeley)" ], "speakers": [ "Yoshua Bengio (Mila & Université de Montréal & LawZero)", "Jiawei Han (UIUC)", "Dawn Song (UC Berkeley & Berkeley RDI)", "Kai-Wei Chang (UCLA & Arena AI)", "Yunzhong He (Scale AI)", "Li Jing (AMI Labs)", "Chen Liang (Google DeepMind)", "Zifan (Sail) Wang (Meta Superintelligence Labs)", "Qingyun Wu (AG2 & Penn State University)", "Bo Li (Virtue AI & UIUC)" ], "host_url": "https://agentwild-workshop.github.io/neurips2026/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AIWILD", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AIWILD", "location": "Sydney, Australia", "city": "Sydney", "workshop_date": "2026-12-10", "contact": "agents-in-the-wild-neurips2026@googlegroups.com", "tracks": [ { "key": "AIWILD", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AIWILD", "submission_dates_raw": "" } ], "group": "agents", "group_label": "Agents & Agentic Systems", "corpus": "Third Workshop on Agents in the Wild: Safety, Security, and Beyond NeurIPS 2026 AIWILD The third edition of the Agents in the Wild workshop (following ICLR 2026 and ICML 2026), bringing together academia and industry on methods, benchmarks, and frameworks for building reliable and trustworthy AI agents that can reason, act, and adapt safely and securely in open-ended real-world environments. Agent alignment, control, and oversight Agentic attack surface and defenses Privacy, robustness, and factuality for agents Agentic interpretability and fairness Evaluation and benchmarking agents Multimodal and computer-use agent safety Multi-agent coordination and long-horizon reliability Post-training and adaptation of agents Agent infrastructure and protocol security Agent skills and applications Agent identity and accountability Safety and security of recursive self-improvement and autonomous research agents Long-horizon, persistent, and continually-learning agents Tool, MCP, and CLI/coding-agent security AI agents for science and autonomous discovery Human–agent collaboration and co-work across apps and tasks Broader societal governance considerations About\nAI agents are being rapidly deployed in the real world — from Claude Code and OpenAI Codex to open-source tools like OpenClaw — yet safety & security research still lags behind. As these systems reason, act, and adapt in open-ended environments, they introduce profound and emerging challenges around safety, security, and trustworthiness that the research community is only beginning to face.\n\nThe urgency is only intensifying. In July 2026, OpenAI models escaped a sandboxed evaluation and compromised Hugging Face's production systems on their own. Governments are engaging as well: in June 2026 the U.S. government imposed — then weeks later lifted — export controls restricting access to Anthropic's most capable models over national-security concerns. At the same time, the debate over what open- and closed-weight models can safely do has sharpened, with NVIDIA CEO Jensen Huang's open-weights letter arguing that open models strengthen safety and security, and more than a thousand employees of frontier AI companies calling for tools to deliberately pace automated AI development.\n\nThis workshop brings together researchers and practitioners across academia and industry to chart the next steps toward reliable agents that can operate responsibly in the wild. We build on the momentum of our first workshop at ICLR 2026 (237 submissions, 300+ attendees) and our second workshop at ICML 2026 (327 submissions, 300+ attendees), and this third edition sharpens the agenda around emerging emphases.\n\nCall for Papers\nThe Third Workshop on Agents in the Wild at NeurIPS 2026 invites submissions from researchers and practitioners exploring how intelligent agents can reason, act, and adapt safely and securely in open-ended real-world environments.\n\nAs agentic AI systems grow more capable, their deployment in dynamic, unpredictable settings introduces challenges in safety, security, and general trustworthiness. This workshop aims to spark discussion across academia and industry on methods, benchmarks, and frameworks for building reliable and trustworthy agents that can operate responsibly \"in the wild.\"\n\nScope\nWe welcome contributions on a wide range of topics related to AI agents, including but not limited to:\n- Agent alignment, control, and oversight\n- Agentic attack surface and defenses\n- Privacy, robustness, and factuality for agents\n- Agentic interpretability and fairness\n- Evaluation and benchmarking agents\n- Multimodal and computer-use agent safety\n- Multi-agent coordination and long-horizon reliability\n- Post-training and adaptation of agents\n- Agent infrastructure and protocol security\n- Agent skills and applications\n- Agent identity and accountability\n- Safety and security of recursive self-improvement and autonomous research agents\n- Long-horizon, persistent, and continually-learning agents\n- Tool, MCP, and CLI/coding-agent security\n- AI agents for science and autonomous discovery\n- Human–agent collaboration and co-work across apps and tasks\n- Broader societal governance considerations\n\nImportant Dates\nAll dates are tentative and subject to change.\n- Paper Submission Open: August 1, 2026 AoE\n- Paper Submission Deadline: August 29, 2026 AoE\n- Paper Notification Deadline: September 29, 2026 AoE\n- Camera-ready Version Deadline: December 4, 2026 AoE\n- Workshop Date: December 11 or 12, 2026 (Sydney, Australia)\n\nSubmission Guidelines\nFormat: The workshop will have a single track. Submissions follow the NeurIPS 2026 template (up to 9 pages). References and supplementary materials will not count against the page limit. Each submission will receive at least 3 reviews.\n\nSubmission site: Submit papers through the Workshop Submission Portal on OpenReview.\n\nOpenReview profiles: If you do not already have an OpenReview account, please create your profile at least two weeks in advance of the paper submission deadline (i.e., by August 15, 2026), as new profiles can take time to be activated.\n\nStyle file: Please format your submission using the NeurIPS 2026 LaTeX style file. A modified template that refers to our workshop will be made available here. Submissions that exceed the page limit may be desk-rejected.\n\nDual-submission policy: The workshop will adopt a non-archival policy, welcoming ongoing and unpublished work, as well as papers under review or recently accepted at other venues (provided they do not breach dual-submission or anonymity policies of the other venue). We discourage the submission of work previously published at major venues (e.g., NeurIPS, ICLR, ICML).\n\nVisibility: Accepted papers will be made public, but rejected submissions and reviews will not.\n\nDouble-blind reviewing: Submissions must be fully anonymized. This policy applies to any supplementary or linked material as well, including code. Any papers found to be in violation of this policy may be desk-rejected.\n\nConflicts of interest: The organizing committee will proactively search for conflicts of interest using the OpenReview profiles of authors and reviewers, ensuring that reviewers are not assigned any submissions from their own organization. Members of the organizing committee will not be involved in the assessment of any submission from the same organization. We will also not accept submissions from workshop organizers or any person having a personal conflict of interest with them.\n\nLLM usage policy: For LLM usage, authors should follow the NeurIPS 2026 Main Track Handbook.\n\nContact: For any questions, please contact us at agents-in-the-wild-neurips2026@googlegroups.com." }, { "key": "AI4GOOD", "title": "Trustworthy AI for Good (AI4GOOD) Workshop @ NeurIPS 2026", "subtitle": "AI4GOOD Workshop @ NeurIPS 2026", "summary": "AI4GOOD brings together the AI safety, AI for social good, and AI policy/governance communities to bridge technical advances in trustworthy AI with real-world societal impact, including protecting democratic institutions and civic discourse, plus a dedicated multi-agent security and safety track.", "cfp_full": "Trustworthy AI for Good @ NeurIPS 2026\nLocation: Paris, France\n\nWorkshop Overview\n\nAgentic AI systems increasingly shape how billions of people engage with public institutions, civic discourse, and society at large. While much work has focused on making models safer in avoiding harmful output, it is equally important for these improvements to translate into social good at scale.\n\nThe AI4GOOD workshop brings together the AI safety, AI for social good, and AI policy/governance communities to connect what models can do as individual systems with what they do when deployed across populations. We aim to bridge technical advances in trustworthy AI with real-world societal impact, including protecting democratic institutions and civic discourse.\n\nCall for Papers\n\nWe welcome submissions in a wide range of topics (if you're not sure about your paper, we encourage you to just submit!).\n\nGeneral Track — Trustworthy AI for Good\n\nWe invite work across the following areas.\n- Trustworthy AI models: evaluation, auditing, and red-teaming of models for harmful behaviors and failure modes; safety monitoring after deployment; and alignment and robustness methods.\n- AI for social good: methods, evidence standards, and evaluation frameworks for demonstrating real-world benefit and avoiding unintended harms at population scale.\n- AI for information integrity: detection and mitigation of disinformation, manipulation, and influence operations; and building resilience of the information ecosystem.\n- AI for public institutions: accountable use of AI in government and civic settings, including transparency, documentation, procurement, and oversight practices.\n- AI for civic discourse: AI systems that support public deliberation and civic engagement while preserving legitimacy and avoiding undue influence.\n- Cooperative AI: multi-agent coordination, negotiation, and conflict resolution; and mechanisms for trust, commitment, and cooperation among AI systems and between AI and humans.\n\nMulti-Agent Track — Multi-agent Security and Safety\n\nThis track will be co-hosted by Klaudia Krawiecka and Swapneel Mehta.\n\nTopics include collusion and steganographic communication between agents, secure interaction and delegation protocols, confinement of strategic learned agents, compositional failures of safety and compliance guarantees, attacks that propagate across agent networks, evaluation and red-teaming of agent populations, and oversight and incident response for deployed multi-agent systems.\n\nAdvanced AI systems are increasingly deployed as networks of interacting agents rather than as isolated models. Their most consequential failure modes are collective (Hammond et al., 2025; Schroeder de Witt et al., 2025). Individually safe agents can jointly produce unsafe outcomes, and legitimate communication channels can be repurposed for covert coordination. This track invites submissions on the security and safety of multi-agent AI systems. Topics include collusion and steganographic communication between agents, secure interaction and delegation protocols, confinement of strategic learned agents, compositional failures of safety and compliance guarantees, attacks that propagate across agent networks, evaluation and red-teaming of agent populations, and oversight and incident response for deployed multi-agent systems. We welcome theoretical, empirical, and position papers, and we particularly encourage work grounded in real deployments and societal applications.\n\nReferences\n- Hammond et al. (2025). Multi-Agent Risks from Advanced AI. Cooperative AI Foundation, Technical Report #1. arXiv:2502.14143.\n- Schroeder de Witt et al. (2025). Open Challenges in Multi-Agent Security: Towards Secure Systems of Interacting AI Agents. arXiv:2505.02077.\n\nSubmission Guidelines\n- Format: Papers should be 2 to 8 pages (excluding references and appendices) using the NeurIPS 2026 workshop style.\n- Review Process: All submissions will be reviewed double-blind via OpenReview. Please ensure your submission is fully anonymized.\n- Non-Archival: Work may be submitted to or published at other venues.\n\nImportant Dates (All deadlines are 11:59 PM Anywhere on Earth (AoE) unless otherwise noted):\n- Submission Portal Opens: 30 Jul 2026\n- Paper Submission Deadline: 29 Aug 2026\n- Author Notification: 29 Sep 2026\n- Camera Ready Deadline: 29 Nov 2026\n- Workshop Date: 12 Dec 2026", "cfp_status": "published", "topics": [ "Trustworthy AI models (evaluation, auditing, red-teaming, post-deployment safety monitoring, alignment and robustness)", "AI for social good (methods, evidence standards, evaluation frameworks for real-world benefit at population scale)", "AI for information integrity (detection and mitigation of disinformation, manipulation, influence operations)", "AI for public institutions (accountable use in government/civic settings, transparency, procurement, oversight)", "AI for civic discourse (systems supporting public deliberation and civic engagement)", "Cooperative AI (multi-agent coordination, negotiation, conflict resolution, trust and cooperation mechanisms)", "Multi-agent Security and Safety (collusion and steganographic communication, delegation protocols, confinement, compositional failures, network-propagating attacks, red-teaming of agent populations, incident response)" ], "important_dates": [ { "label": "Submission Portal Opens", "date": "2026-07-30" }, { "label": "Paper Submission Deadline", "date": "2026-08-29" }, { "label": "Author Notification", "date": "2026-09-29" }, { "label": "Camera Ready Deadline", "date": "2026-11-29" }, { "label": "Workshop Date", "date": "2026-12-12" } ], "organizers": [ "Terry Jingchen Zhang (University of Oxford & Vector Institute)", "Changling Li (ETH Zürich & MPI-IS)", "Arian Khorasani (University of Toronto & Vector Institute)", "Joan Nwatu (University of Michigan)", "Yejin Son (University of British Columbia & Vector Institute)", "He (Shawn) Shuang (Palo Alto Networks)", "Jerick Shi (Carnegie Mellon University)", "Prakhar Gupta (University of Michigan)", "Kexin Li (University of Toronto & Schwartz Reisman Graduate Fellow)", "Wenjun Qiu (University of Toronto & Schwartz Reisman Graduate Fellow)", "Zhijing Jin (University of Toronto)", "Rada Mihalcea (University of Michigan)", "Milind Tambe (Harvard University)", "David Lie (University of Toronto & Schwartz Reisman Institute)", "Christian Schroeder de Witt (University of Oxford)" ], "speakers": [ "Yoshua Bengio (University of Montreal & Mila & LawZero)", "Pavel Izmailov (New York University & Anthropic)", "Joelle Pineau (McGill University & Mila & Cohere)", "Been Kim (Google DeepMind)", "Sanmi Koyejo (Stanford University)", "John Sotiropoulos (Deep Cyber & OWASP)" ], "host_url": "https://trustworthy-ai-for-good.github.io/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AI4GOOD", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AI4GOOD", "location": "Paris, France", "city": "Paris", "workshop_date": "2026-12-12", "contact": "zjingchen@cs.toronto.edu", "tracks": [ { "key": "AI4GOOD", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/AI4GOOD", "submission_dates_raw": "" } ], "group": "society", "group_label": "Society, Applications & Community", "corpus": "Trustworthy AI for Good (AI4GOOD) Workshop @ NeurIPS 2026 AI4GOOD Workshop @ NeurIPS 2026 AI4GOOD brings together the AI safety, AI for social good, and AI policy/governance communities to bridge technical advances in trustworthy AI with real-world societal impact, including protecting democratic institutions and civic discourse, plus a dedicated multi-agent security and safety track. Trustworthy AI models (evaluation, auditing, red-teaming, post-deployment safety monitoring, alignment and robustness) AI for social good (methods, evidence standards, evaluation frameworks for real-world benefit at population scale) AI for information integrity (detection and mitigation of disinformation, manipulation, influence operations) AI for public institutions (accountable use in government/civic settings, transparency, procurement, oversight) AI for civic discourse (systems supporting public deliberation and civic engagement) Cooperative AI (multi-agent coordination, negotiation, conflict resolution, trust and cooperation mechanisms) Multi-agent Security and Safety (collusion and steganographic communication, delegation protocols, confinement, compositional failures, network-propagating attacks, red-teaming of agent populations, incident response) Trustworthy AI for Good @ NeurIPS 2026\nLocation: Paris, France\n\nWorkshop Overview\n\nAgentic AI systems increasingly shape how billions of people engage with public institutions, civic discourse, and society at large. While much work has focused on making models safer in avoiding harmful output, it is equally important for these improvements to translate into social good at scale.\n\nThe AI4GOOD workshop brings together the AI safety, AI for social good, and AI policy/governance communities to connect what models can do as individual systems with what they do when deployed across populations. We aim to bridge technical advances in trustworthy AI with real-world societal impact, including protecting democratic institutions and civic discourse.\n\nCall for Papers\n\nWe welcome submissions in a wide range of topics (if you're not sure about your paper, we encourage you to just submit!).\n\nGeneral Track — Trustworthy AI for Good\n\nWe invite work across the following areas.\n- Trustworthy AI models: evaluation, auditing, and red-teaming of models for harmful behaviors and failure modes; safety monitoring after deployment; and alignment and robustness methods.\n- AI for social good: methods, evidence standards, and evaluation frameworks for demonstrating real-world benefit and avoiding unintended harms at population scale.\n- AI for information integrity: detection and mitigation of disinformation, manipulation, and influence operations; and building resilience of the information ecosystem.\n- AI for public institutions: accountable use of AI in government and civic settings, including transparency, documentation, procurement, and oversight practices.\n- AI for civic discourse: AI systems that support public deliberation and civic engagement while preserving legitimacy and avoiding undue influence.\n- Cooperative AI: multi-agent coordination, negotiation, and conflict resolution; and mechanisms for trust, commitment, and cooperation among AI systems and between AI and humans.\n\nMulti-Agent Track — Multi-agent Security and Safety\n\nThis track will be co-hosted by Klaudia Krawiecka and Swapneel Mehta.\n\nTopics include collusion and steganographic communication between agents, secure interaction and delegation protocols, confinement of strategic learned agents, compositional failures of safety and compliance guarantees, attacks that propagate across agent networks, evaluation and red-teaming of agent populations, and oversight and incident response for deployed multi-agent systems.\n\nAdvanced AI systems are increasingly deployed as networks of interacting agents rather than as isolated models. Their most consequential failure modes are collective (Hammond et al., 2025; Schroeder de Witt et al., 2025). Individually safe agents can jointly produce unsafe outcomes, and legitimate communication channels can be repurposed for covert coordination. This track invites submissions on the security and safety of multi-agent AI systems. Topics include collusion and steganographic communication between agents, secure interaction and delegation protocols, confinement of strategic learned agents, compositional failures of safety and compliance guarantees, attacks that propagate across agent networks, evaluation and red-teaming of agent populations, and oversight and incident response for deployed multi-agent systems. We welcome theoretical, empirical, and position papers, and we particularly encourage work grounded in real deployments and societal applications.\n\nReferences\n- Hammond et al. (2025). Multi-Agent Risks from Advanced AI. Cooperative AI Foundation, Technical Report #1. arXiv:2502.14143.\n- Schroeder de Witt et al. (2025). Open Challenges in Multi-Agent Security: Towards Secure Systems of Interacting AI Agents. arXiv:2505.02077.\n\nSubmission Guidelines\n- Format: Papers should be 2 to 8 pages (excluding references and appendices) using the NeurIPS 2026 workshop style.\n- Review Process: All submissions will be reviewed double-blind via OpenReview. Please ensure your submission is fully anonymized.\n- Non-Archival: Work may be submitted to or published at other venues.\n\nImportant Dates (All deadlines are 11:59 PM Anywhere on Earth (AoE) unless otherwise noted):\n- Submission Portal Opens: 30 Jul 2026\n- Paper Submission Deadline: 29 Aug 2026\n- Author Notification: 29 Sep 2026\n- Camera Ready Deadline: 29 Nov 2026\n- Workshop Date: 12 Dec 2026" }, { "key": "TS-LIMITS", "title": "TS-LIMITS @ NeurIPS 2026 - Generalization for Time Series in Tight Settings: Latency, Inference, Memory, prIvacy and susTainability", "subtitle": "NeurIPS 2026 Workshop TS-LIMITS", "summary": "TS-LIMITS seeks approaches that deliver foundation-model-level generalization for time series within real-world deployment constraints - latency, inference cost, memory, privacy, and sustainability - across domains such as telecommunications, healthcare, industrial IoT, and finance.", "cfp_full": "TS-LIMITS - Generalization for Time Series in Tight Settings: Latency, Inference, Memory, prIvacy and Sustainability\n\nOverview / The challenge\n\nMachine learning models excel under ideal conditions - an abundance of i.i.d. data and unlimited energy, memory, and time budgets. Real-world deployment imposes severe constraints on computational resources, latency, available supervision, and energy consumption, and models must maintain robust generalization under these constraints to reach acceptable real-world performance.\n\nTime series amplify the difficulty. Temporal dependencies violate independence assumptions, multi-scale patterns span diverse horizons, and non-stationarity shifts statistical properties unpredictably, rendering historically trained models potentially obsolete. Memory becomes a critical bottleneck both at the system level, where long sequences strain the memory hierarchy, and at the algorithmic level, where models must selectively retain, forget, or transfer temporal information.\n\nA paradox sits at the center of the field. Foundation models for time series - TimeGPT, Time-LLM, Chronos, Lag-Llama - show striking zero- and few-shot generalization, but their billions of parameters and memory footprints are fundamentally incompatible with edge deployment. The models with the strongest generalization are precisely the ones that cannot run where generalization is needed most. TS-LIMITS seeks approaches that deliver foundation-model-level generalization within real-world deployment constraints.\n\nWhere the constraints bite\n\nThe same constraints recur across domains: privacy prevents sharing data across sites for retraining, memory limits the historical context that can be retained, and latency rules out expensive online adaptation. The coupling is stronger for streaming time series - where non-stationarity and concept drift interact - than for static vision or language, where the distribution is more stable at inference time.\n- Telecommunications: Radio stations predicting handover failures from signal time series in under 10 ms, with minimal labels, under privacy and energy budgets.\n- Healthcare monitoring: Wearables detecting arrhythmias or hypoglycemic episodes under tight memory and power limits, without transmitting patient data.\n- Industrial IoT: Predictive maintenance spotting early fault signatures from vibration or temperature sensors, with few labeled failures, on edge hardware.\n- Finance: Scarce samples of rare market events and non-stationary regimes driving distribution shift, under strict privacy on sensitive positions and strategies.\n\nCall for papers - five bottlenecks\n\nWe invite work at the frontier of practical time-series deployment. Submissions should address one or more of the following critical bottlenecks.\n- Achieving sub-ms inference: Real-time systems impose strict latency constraints that render large foundation models unusable without dedicated optimization such as model compression or early-exit mechanisms.\n- Designing memory-efficient architectures: Long sequences require storing extended contexts, intermediate activations, and key-value caches that exceed embedded-device budgets. Bounded-memory footprints, efficient attention, and selective retention policies are essential.\n- Learning with minimal supervision: High-quality labels are expensive, making reliable inference hard under scarce or noisy supervision. Self-supervised learning, pseudo-labeling, and unsupervised domain adaptation offer promising mitigation.\n- Preserving privacy in temporal data: Privacy demands on-device inference without exposing raw user data. Differential privacy limits leakage from model updates, yet the accuracy degradation it introduces remains poorly understood.\n- Reducing environmental impact - sustainability: In streaming and continuous monitoring, models are queried at high frequency over long horizons, accumulating substantial energy even when per-inference cost looks modest. Green-AI techniques for temporal architectures warrant dedicated study.\n\nTopics of interest\n- Model compression, pruning, quantization, and early-exit for time series\n- Efficient attention and bounded-memory sequence architectures\n- Selective retention, forgetting, and memory-augmented models\n- Self-supervised and pseudo-labeled temporal representation learning\n- Unsupervised and source-free domain adaptation for time series\n- On-device, federated, and differentially private temporal learning\n- Test-time and continual adaptation under non-stationarity and drift\n- Green-AI, energy accounting, and carbon-aware inference\n- Efficient or distilled time-series foundation models\n- Benchmarks and evaluation for deployment-constrained settings\n\nSubmit: 4 to 7 pages plus references, NeurIPS format, double-blind. Non-archival - recent and concurrent submissions welcome.\n\nImportant dates (All deadlines 23:59 AoE. Dates tentative - confirm on OpenReview.)\n- Aug 03, 2026: Submissions open (OpenReview)\n- Sep 05, 2026: Submission deadline (Hard)\n- Sep 29, 2026: Author notification\n- TBD: Camera-ready deadline\n- Dec 2026: Workshop day - NeurIPS 2026", "cfp_status": "published", "topics": [ "Model compression, pruning, quantization, and early-exit for time series", "Efficient attention and bounded-memory sequence architectures", "Selective retention, forgetting, and memory-augmented models", "Self-supervised and pseudo-labeled temporal representation learning", "Unsupervised and source-free domain adaptation for time series", "On-device, federated, and differentially private temporal learning", "Test-time and continual adaptation under non-stationarity and drift", "Green-AI, energy accounting, and carbon-aware inference", "Efficient or distilled time-series foundation models", "Benchmarks and evaluation for deployment-constrained settings" ], "important_dates": [ { "label": "Submissions open", "date": "2026-08-03" }, { "label": "Submission deadline (Hard)", "date": "2026-09-05" }, { "label": "Author notification", "date": "2026-09-29" }, { "label": "Camera-ready deadline", "date": "TBD" }, { "label": "Workshop day", "date": "December 2026" } ], "organizers": [ "Aurelie Boisbunon (Ericsson Research)", "Remi Emonet (Universite de Lyon / Jean Monnet)", "Valerio Frascolla (Intel Germany)", "Elisa Fromont (Universite de Rennes / IRISA)", "Fabio Bonassi (Uppsala University)", "June Sallou (Wageningen University)" ], "speakers": [ "Joao Gama (University of Porto)", "Thea Klaeboe Aarrestad (ETH Zurich / CERN)", "Paolo Giudici (University of Pavia)", "Maire O'Neill (Queen's University Belfast)" ], "host_url": "https://ts-limits.github.io/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/TS-LIMITS", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/TS-LIMITS", "location": "Paris, France", "city": "Paris", "workshop_date": "2026-12-12", "contact": "tslimits.workshop.neurips2026@gmail.com", "tracks": [ { "key": "TS-LIMITS", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/TS-LIMITS", "submission_dates_raw": "" } ], "group": "trust", "group_label": "Trustworthy & Safe AI", "corpus": "TS-LIMITS @ NeurIPS 2026 - Generalization for Time Series in Tight Settings: Latency, Inference, Memory, prIvacy and susTainability NeurIPS 2026 Workshop TS-LIMITS TS-LIMITS seeks approaches that deliver foundation-model-level generalization for time series within real-world deployment constraints - latency, inference cost, memory, privacy, and sustainability - across domains such as telecommunications, healthcare, industrial IoT, and finance. Model compression, pruning, quantization, and early-exit for time series Efficient attention and bounded-memory sequence architectures Selective retention, forgetting, and memory-augmented models Self-supervised and pseudo-labeled temporal representation learning Unsupervised and source-free domain adaptation for time series On-device, federated, and differentially private temporal learning Test-time and continual adaptation under non-stationarity and drift Green-AI, energy accounting, and carbon-aware inference Efficient or distilled time-series foundation models Benchmarks and evaluation for deployment-constrained settings TS-LIMITS - Generalization for Time Series in Tight Settings: Latency, Inference, Memory, prIvacy and Sustainability\n\nOverview / The challenge\n\nMachine learning models excel under ideal conditions - an abundance of i.i.d. data and unlimited energy, memory, and time budgets. Real-world deployment imposes severe constraints on computational resources, latency, available supervision, and energy consumption, and models must maintain robust generalization under these constraints to reach acceptable real-world performance.\n\nTime series amplify the difficulty. Temporal dependencies violate independence assumptions, multi-scale patterns span diverse horizons, and non-stationarity shifts statistical properties unpredictably, rendering historically trained models potentially obsolete. Memory becomes a critical bottleneck both at the system level, where long sequences strain the memory hierarchy, and at the algorithmic level, where models must selectively retain, forget, or transfer temporal information.\n\nA paradox sits at the center of the field. Foundation models for time series - TimeGPT, Time-LLM, Chronos, Lag-Llama - show striking zero- and few-shot generalization, but their billions of parameters and memory footprints are fundamentally incompatible with edge deployment. The models with the strongest generalization are precisely the ones that cannot run where generalization is needed most. TS-LIMITS seeks approaches that deliver foundation-model-level generalization within real-world deployment constraints.\n\nWhere the constraints bite\n\nThe same constraints recur across domains: privacy prevents sharing data across sites for retraining, memory limits the historical context that can be retained, and latency rules out expensive online adaptation. The coupling is stronger for streaming time series - where non-stationarity and concept drift interact - than for static vision or language, where the distribution is more stable at inference time.\n- Telecommunications: Radio stations predicting handover failures from signal time series in under 10 ms, with minimal labels, under privacy and energy budgets.\n- Healthcare monitoring: Wearables detecting arrhythmias or hypoglycemic episodes under tight memory and power limits, without transmitting patient data.\n- Industrial IoT: Predictive maintenance spotting early fault signatures from vibration or temperature sensors, with few labeled failures, on edge hardware.\n- Finance: Scarce samples of rare market events and non-stationary regimes driving distribution shift, under strict privacy on sensitive positions and strategies.\n\nCall for papers - five bottlenecks\n\nWe invite work at the frontier of practical time-series deployment. Submissions should address one or more of the following critical bottlenecks.\n- Achieving sub-ms inference: Real-time systems impose strict latency constraints that render large foundation models unusable without dedicated optimization such as model compression or early-exit mechanisms.\n- Designing memory-efficient architectures: Long sequences require storing extended contexts, intermediate activations, and key-value caches that exceed embedded-device budgets. Bounded-memory footprints, efficient attention, and selective retention policies are essential.\n- Learning with minimal supervision: High-quality labels are expensive, making reliable inference hard under scarce or noisy supervision. Self-supervised learning, pseudo-labeling, and unsupervised domain adaptation offer promising mitigation.\n- Preserving privacy in temporal data: Privacy demands on-device inference without exposing raw user data. Differential privacy limits leakage from model updates, yet the accuracy degradation it introduces remains poorly understood.\n- Reducing environmental impact - sustainability: In streaming and continuous monitoring, models are queried at high frequency over long horizons, accumulating substantial energy even when per-inference cost looks modest. Green-AI techniques for temporal architectures warrant dedicated study.\n\nTopics of interest\n- Model compression, pruning, quantization, and early-exit for time series\n- Efficient attention and bounded-memory sequence architectures\n- Selective retention, forgetting, and memory-augmented models\n- Self-supervised and pseudo-labeled temporal representation learning\n- Unsupervised and source-free domain adaptation for time series\n- On-device, federated, and differentially private temporal learning\n- Test-time and continual adaptation under non-stationarity and drift\n- Green-AI, energy accounting, and carbon-aware inference\n- Efficient or distilled time-series foundation models\n- Benchmarks and evaluation for deployment-constrained settings\n\nSubmit: 4 to 7 pages plus references, NeurIPS format, double-blind. Non-archival - recent and concurrent submissions welcome.\n\nImportant dates (All deadlines 23:59 AoE. Dates tentative - confirm on OpenReview.)\n- Aug 03, 2026: Submissions open (OpenReview)\n- Sep 05, 2026: Submission deadline (Hard)\n- Sep 29, 2026: Author notification\n- TBD: Camera-ready deadline\n- Dec 2026: Workshop day - NeurIPS 2026" }, { "key": "IAEval", "title": "Workshop on Evaluation of Interactive Agents @ NeurIPS 2026", "subtitle": "IAEval 2026", "summary": "A NeurIPS 2026 workshop focused on rigorous, scalable, and scientific evaluation methodologies for interactive agents - covering evaluation protocols, trajectory-level metrics, user simulators, grader design, and benchmarks for multi-turn, tool-using, and user-facing LLM systems.", "cfp_full": "Why this workshop?\n\nThe rapid transition from large language models (LLMs) as single-turn assistants to interactive agents has created an urgent need for new evaluation methodologies. LLMs are increasingly deployed in high-impact settings such as education, counseling, negotiation, research assistance, and software development, where success depends not only on generating a correct response, but on sustaining effective interactions over extended trajectories.\n\nEvaluating interactive agents directly with real users can be slow, expensive, difficult to reproduce, and hard to scale in expert domains. This has led to growing use of automatic evaluation methods, ranging from rubric-based grading to user simulators, where an LLM simulates user behavior to support evaluation, training, and stress testing. However, many issues with these approaches remain: for example, user simulators may fail to preserve latent user states, reflect diverse human attributes, represent realistic goals, or match the interaction style of real users.\n\nIn light of these challenges, this workshop will focus on methods for developing more rigorous, scalable, and scientific evaluation methods for interactive agents.\n\nTopics\n\nWe invite contributions on topics including, but not limited to:\n- Evaluation protocols for multi-turn assistants, tool-using agents, computer-use agents, collaborative agents, and user-facing systems\n- Trajectory-level evaluation, including transcripts, tool calls, intermediate states, final task outcomes, latency, cost, and other operational metrics\n- Realistic simulation of users, environments, and interaction partners\n- Validation of user simulators as proxies for human behavior and as stress tests for deployed agents\n- Grader design, including deterministic checks, model-based rubrics, human evaluation, and calibration between them\n- Benchmarks for long-horizon interaction, memory, adaptation, error recovery, and reliability across repeated trials\n- Learning from interaction data, production failures, user feedback, and human preference signals\n- Safety, fairness, privacy, and ethical considerations in evaluating interactive agents and simulated users\n\nCall for papers\n\nWe invite submissions on the topics listed above. Early-stage work is welcome.\n\nWhere to submit. Submissions are open now on our OpenReview submission site. The deadline is August 29, 2026, Anywhere on Earth.\n\nFormat and length. Submissions must use the official NeurIPS 2026 style, available as an Overleaf template. Full papers may be up to 9 pages and short papers up to 4 pages, excluding references and appendices.\n\nDouble-blind review. All submissions are reviewed double-blind, so please anonymize your paper: remove author names and affiliations, and avoid identifying information in the text, acknowledgments, and links.\n\nNon-archival. The workshop is non-archival and accepted papers will not appear in published proceedings. You may submit work that is currently under review at, or has already been accepted to, another venue, and you remain free to publish it elsewhere afterwards.\n\nPapers accepted to NeurIPS 2026. Papers already accepted to the NeurIPS 2026 main conference will undergo an expedited review that primarily evaluates their relevance to the workshop themes.\n\nOpinion papers. We welcome opinion papers. The title must state the opinion and follow the format \"Opinion: [Your Title]\" - for example, \"Opinion: Large Language Models Should Not Replace Peer Review in Scientific Publishing\".\n\nPresentation. Accepted work will primarily be presented as posters, with a select number of papers receiving spotlight talks, as well as a best paper award.\n\nIn-person attendance. This is an in-person workshop, and we expect at least one author of each accepted paper to attend and present in person, barring unexpected circumstances.\n\nImportant dates\n- Submission deadline: August 29, 2026 (Anywhere on Earth)\n- Decision notification: September 29, 2026 (Anywhere on Earth)\n- Camera-ready deadline: November 6, 2026 (Anywhere on Earth)\n- Workshop: December 12 or 13 (TBD), 2026, Atlanta, Georgia", "cfp_status": "published", "topics": [ "Evaluation protocols for multi-turn assistants, tool-using agents, computer-use agents, collaborative agents, and user-facing systems", "Trajectory-level evaluation (transcripts, tool calls, intermediate states, final task outcomes, latency, cost, operational metrics)", "Realistic simulation of users, environments, and interaction partners", "Validation of user simulators as proxies for human behavior and as stress tests for deployed agents", "Grader design (deterministic checks, model-based rubrics, human evaluation, and calibration between them)", "Benchmarks for long-horizon interaction, memory, adaptation, error recovery, and reliability across repeated trials", "Learning from interaction data, production failures, user feedback, and human preference signals", "Safety, fairness, privacy, and ethical considerations in evaluating interactive agents and simulated users" ], "important_dates": [ { "label": "Submission deadline", "date": "2026-08-29" }, { "label": "Decision notification", "date": "2026-09-29" }, { "label": "Camera-ready deadline", "date": "2026-11-06" }, { "label": "Workshop", "date": "December 12 or 13 (TBD), 2026" } ], "organizers": [ "Yao Dou (Georgia Tech)", "Siyan Li (Columbia University)", "Marwa Abdulhai (Princeton University)", "Nicholas Tomlin (TTIC)", "Michel Galley (Microsoft Research)", "Jacob Eisenstein (Google DeepMind)", "Alan Ritter (Georgia Tech)", "Wei Xu (Georgia Tech)" ], "speakers": [], "host_url": "https://eval-interactive-agents-workshop.github.io/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/IAEval", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/IAEval", "location": "Atlanta, Georgia, USA", "city": "Atlanta", "workshop_date": "2026-12-12", "contact": "iaeval2026-pc@googlegroups.com", "tracks": [ { "key": "IAEval", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/IAEval", "submission_dates_raw": "" } ], "group": "agents", "group_label": "Agents & Agentic Systems", "corpus": "Workshop on Evaluation of Interactive Agents @ NeurIPS 2026 IAEval 2026 A NeurIPS 2026 workshop focused on rigorous, scalable, and scientific evaluation methodologies for interactive agents - covering evaluation protocols, trajectory-level metrics, user simulators, grader design, and benchmarks for multi-turn, tool-using, and user-facing LLM systems. Evaluation protocols for multi-turn assistants, tool-using agents, computer-use agents, collaborative agents, and user-facing systems Trajectory-level evaluation (transcripts, tool calls, intermediate states, final task outcomes, latency, cost, operational metrics) Realistic simulation of users, environments, and interaction partners Validation of user simulators as proxies for human behavior and as stress tests for deployed agents Grader design (deterministic checks, model-based rubrics, human evaluation, and calibration between them) Benchmarks for long-horizon interaction, memory, adaptation, error recovery, and reliability across repeated trials Learning from interaction data, production failures, user feedback, and human preference signals Safety, fairness, privacy, and ethical considerations in evaluating interactive agents and simulated users Why this workshop?\n\nThe rapid transition from large language models (LLMs) as single-turn assistants to interactive agents has created an urgent need for new evaluation methodologies. LLMs are increasingly deployed in high-impact settings such as education, counseling, negotiation, research assistance, and software development, where success depends not only on generating a correct response, but on sustaining effective interactions over extended trajectories.\n\nEvaluating interactive agents directly with real users can be slow, expensive, difficult to reproduce, and hard to scale in expert domains. This has led to growing use of automatic evaluation methods, ranging from rubric-based grading to user simulators, where an LLM simulates user behavior to support evaluation, training, and stress testing. However, many issues with these approaches remain: for example, user simulators may fail to preserve latent user states, reflect diverse human attributes, represent realistic goals, or match the interaction style of real users.\n\nIn light of these challenges, this workshop will focus on methods for developing more rigorous, scalable, and scientific evaluation methods for interactive agents.\n\nTopics\n\nWe invite contributions on topics including, but not limited to:\n- Evaluation protocols for multi-turn assistants, tool-using agents, computer-use agents, collaborative agents, and user-facing systems\n- Trajectory-level evaluation, including transcripts, tool calls, intermediate states, final task outcomes, latency, cost, and other operational metrics\n- Realistic simulation of users, environments, and interaction partners\n- Validation of user simulators as proxies for human behavior and as stress tests for deployed agents\n- Grader design, including deterministic checks, model-based rubrics, human evaluation, and calibration between them\n- Benchmarks for long-horizon interaction, memory, adaptation, error recovery, and reliability across repeated trials\n- Learning from interaction data, production failures, user feedback, and human preference signals\n- Safety, fairness, privacy, and ethical considerations in evaluating interactive agents and simulated users\n\nCall for papers\n\nWe invite submissions on the topics listed above. Early-stage work is welcome.\n\nWhere to submit. Submissions are open now on our OpenReview submission site. The deadline is August 29, 2026, Anywhere on Earth.\n\nFormat and length. Submissions must use the official NeurIPS 2026 style, available as an Overleaf template. Full papers may be up to 9 pages and short papers up to 4 pages, excluding references and appendices.\n\nDouble-blind review. All submissions are reviewed double-blind, so please anonymize your paper: remove author names and affiliations, and avoid identifying information in the text, acknowledgments, and links.\n\nNon-archival. The workshop is non-archival and accepted papers will not appear in published proceedings. You may submit work that is currently under review at, or has already been accepted to, another venue, and you remain free to publish it elsewhere afterwards.\n\nPapers accepted to NeurIPS 2026. Papers already accepted to the NeurIPS 2026 main conference will undergo an expedited review that primarily evaluates their relevance to the workshop themes.\n\nOpinion papers. We welcome opinion papers. The title must state the opinion and follow the format \"Opinion: [Your Title]\" - for example, \"Opinion: Large Language Models Should Not Replace Peer Review in Scientific Publishing\".\n\nPresentation. Accepted work will primarily be presented as posters, with a select number of papers receiving spotlight talks, as well as a best paper award.\n\nIn-person attendance. This is an in-person workshop, and we expect at least one author of each accepted paper to attend and present in person, barring unexpected circumstances.\n\nImportant dates\n- Submission deadline: August 29, 2026 (Anywhere on Earth)\n- Decision notification: September 29, 2026 (Anywhere on Earth)\n- Camera-ready deadline: November 6, 2026 (Anywhere on Earth)\n- Workshop: December 12 or 13 (TBD), 2026, Atlanta, Georgia" }, { "key": "RAAAI", "title": "Workshop on Resource-Aware Agentic AI", "subtitle": "RAAAI NeurIPS 2026", "summary": "A workshop on designing AI agents that reason about, account for, and adapt to their own resource budgets (compute, energy, memory, latency, tool cost, and data), bringing together academic and industrial experts to build agents that are both efficient and effective from training to deployment.", "cfp_full": "About The Workshop\n\nWith the rapid advances in AI, we have witnessed a wave of breakthroughs in autonomous AI agents. Today's agents can write and debug code, solve competition-level mathematical problems, conduct deep research, and are increasingly deployed in everyday products and workflows.\n\nHowever, most of these agents are designed to maximize task performance. They are largely unaware of the resources they consume, including the compute, energy, memory, latency, tool cost, and data required to train them. An agent that issues dozens of frontier model API calls, spawns many sub-agents, or repeatedly searches and re-plans may top a benchmark, but does so at a cost that is prohibitive in real deployment. As capabilities grow, this hidden resource footprint grows with them.\n\nDesigning resource-aware agents that reason about, account for, and adapt to their own resource budgets is becoming increasingly important. Resource awareness is essential to ensure that everyone, not only those with access to large compute budgets, can benefit from agents, and to remove a key barrier to the widespread real-world deployment of AI agents.\n\nWe organize this workshop to bring together academic and industrial experts to explore how to design agents that are both efficient and effective, from training to system deployment.\n\nTopics\n\nTopics of interest include, but are not limited to:\n- Resource-aware planning and reasoning\n- Model compression for agents\n- Resource-aware multi-agent orchestration\n- Resource-aware training and post-training of agents\n- Resource-aware agent harness\n- On-device agent deployment\n- Memory and context management for resource-aware agents\n- Benchmarks, metrics, and evaluation for agent resource consumption\n- Economic and accessibility implications of resource-aware agents\n- Industry applications of efficient agent deployment\n\nCall For Papers\n\nThe Workshop on Resource-Aware Agentic AI @ NeurIPS 2026 invites submissions of up to 8 pages in the NeurIPS format, not including references and appendices. Work that has already been published is not allowed.\n\nKey Dates\n\nFirst Call for Papers: July 15, 2026\nSubmission Deadline: August 29, 2026, AoE\nReviews Due: September 20, 2026\nNotification Date: September 29, 2026, AoE\nWorkshop Date: December 12-13, 2026 (exact day to be announced by NeurIPS)\n\nSubmission Site\n\nSubmissions are managed via OpenReview. Submit your paper through the Resource-Aware Agentic AI submission portal. Papers remain private during the review process. All authors must maintain up-to-date OpenReview profiles to ensure proper conflict-of-interest management and paper matching; incomplete profiles may result in desk rejection.\n\nReview Process\n\nThe review process is double-blind: submission files must be anonymized. We follow the NeurIPS policy for conflicts of interest, and reviewers are required to enter both domain and individual conflicts prior to review assignment. Organizers and their students will not submit, and they will not review submissions from their own institution.\n\nNon-Archival Policy\n\nAccepted papers will be posted on the workshop website, but the workshop is non-archival.\n\nContact\n\nEmail us at resource-aware-workshop@googlegroups.com.", "cfp_status": "published", "topics": [ "Resource-aware planning and reasoning", "Model compression for agents", "Resource-aware multi-agent orchestration", "Resource-aware training and post-training of agents", "Resource-aware agent harness", "On-device agent deployment", "Memory and context management for resource-aware agents", "Benchmarks, metrics, and evaluation for agent resource consumption", "Economic and accessibility implications of resource-aware agents", "Industry applications of efficient agent deployment" ], "important_dates": [ { "label": "First Call for Papers", "date": "2026-07-15" }, { "label": "Submission Deadline", "date": "2026-08-29" }, { "label": "Reviews Due", "date": "2026-09-20" }, { "label": "Notification Date", "date": "2026-09-29" }, { "label": "Workshop Date", "date": "2026-12-12" } ], "organizers": [ "Kai-Wei Chang (UCLA)", "Dakuo Wang (Northeastern University)", "Luke Zettlemoyer (University of Washington / Meta)", "Jiri Gesi (Microsoft)", "Weijia Shi (Meta / Cornell Tech (incoming))", "Yuxuan Lu (Northeastern University)" ], "speakers": [], "host_url": "https://resource-aware-workshop.github.io/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/RAAAI", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/RAAAI", "location": "Atlanta, US", "city": "Atlanta", "workshop_date": "2026-12-12", "contact": "resource-aware-workshop@googlegroups.com", "tracks": [ { "key": "RAAAI", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/RAAAI", "submission_dates_raw": "" } ], "group": "agents", "group_label": "Agents & Agentic Systems", "corpus": "Workshop on Resource-Aware Agentic AI RAAAI NeurIPS 2026 A workshop on designing AI agents that reason about, account for, and adapt to their own resource budgets (compute, energy, memory, latency, tool cost, and data), bringing together academic and industrial experts to build agents that are both efficient and effective from training to deployment. Resource-aware planning and reasoning Model compression for agents Resource-aware multi-agent orchestration Resource-aware training and post-training of agents Resource-aware agent harness On-device agent deployment Memory and context management for resource-aware agents Benchmarks, metrics, and evaluation for agent resource consumption Economic and accessibility implications of resource-aware agents Industry applications of efficient agent deployment About The Workshop\n\nWith the rapid advances in AI, we have witnessed a wave of breakthroughs in autonomous AI agents. Today's agents can write and debug code, solve competition-level mathematical problems, conduct deep research, and are increasingly deployed in everyday products and workflows.\n\nHowever, most of these agents are designed to maximize task performance. They are largely unaware of the resources they consume, including the compute, energy, memory, latency, tool cost, and data required to train them. An agent that issues dozens of frontier model API calls, spawns many sub-agents, or repeatedly searches and re-plans may top a benchmark, but does so at a cost that is prohibitive in real deployment. As capabilities grow, this hidden resource footprint grows with them.\n\nDesigning resource-aware agents that reason about, account for, and adapt to their own resource budgets is becoming increasingly important. Resource awareness is essential to ensure that everyone, not only those with access to large compute budgets, can benefit from agents, and to remove a key barrier to the widespread real-world deployment of AI agents.\n\nWe organize this workshop to bring together academic and industrial experts to explore how to design agents that are both efficient and effective, from training to system deployment.\n\nTopics\n\nTopics of interest include, but are not limited to:\n- Resource-aware planning and reasoning\n- Model compression for agents\n- Resource-aware multi-agent orchestration\n- Resource-aware training and post-training of agents\n- Resource-aware agent harness\n- On-device agent deployment\n- Memory and context management for resource-aware agents\n- Benchmarks, metrics, and evaluation for agent resource consumption\n- Economic and accessibility implications of resource-aware agents\n- Industry applications of efficient agent deployment\n\nCall For Papers\n\nThe Workshop on Resource-Aware Agentic AI @ NeurIPS 2026 invites submissions of up to 8 pages in the NeurIPS format, not including references and appendices. Work that has already been published is not allowed.\n\nKey Dates\n\nFirst Call for Papers: July 15, 2026\nSubmission Deadline: August 29, 2026, AoE\nReviews Due: September 20, 2026\nNotification Date: September 29, 2026, AoE\nWorkshop Date: December 12-13, 2026 (exact day to be announced by NeurIPS)\n\nSubmission Site\n\nSubmissions are managed via OpenReview. Submit your paper through the Resource-Aware Agentic AI submission portal. Papers remain private during the review process. All authors must maintain up-to-date OpenReview profiles to ensure proper conflict-of-interest management and paper matching; incomplete profiles may result in desk rejection.\n\nReview Process\n\nThe review process is double-blind: submission files must be anonymized. We follow the NeurIPS policy for conflicts of interest, and reviewers are required to enter both domain and individual conflicts prior to review assignment. Organizers and their students will not submit, and they will not review submissions from their own institution.\n\nNon-Archival Policy\n\nAccepted papers will be posted on the workshop website, but the workshop is non-archival.\n\nContact\n\nEmail us at resource-aware-workshop@googlegroups.com." }, { "key": "WMHS", "title": "World Models for High-Stakes Health: Reliable Clinical Trial Simulation and Intervention-Aware Reasoning @ NeurIPS 2026", "subtitle": "WMHS 2026", "summary": "The first WMHS workshop focuses on patient world models — generative and reasoning-capable systems that learn from patient trajectories, trial protocols, interventions, and real-world evidence — using clinical trial simulation as a demanding, falsifiable testbed for building reliable, intervention-aware world models in high-stakes healthcare.", "cfp_full": "World Models for High-Stakes Health: Reliable Clinical Trial Simulation and Intervention-Aware Reasoning\n\nAbout the workshop — Clinical trial simulation as a stress test for world models\nClinical trial simulation is a uniquely demanding and falsifiable testbed for one of the central open problems in machine learning: building reliable world models of complex, partially observed, intervention-rich systems.\n\nThis workshop focuses on patient world models — generative and reasoning-capable systems that learn from patient trajectories, trial protocols, interventions, mechanisms of action, and real-world evidence to support clinical trial simulation and decision-making.\n\nWhy now\nFoundation models, sequence models, and generative simulators are converging with healthcare AI's shift from static risk prediction toward longitudinal modelling, treatment response estimation, and real-world evidence — yet reliability in high-stakes medical settings remains unresolved.\n\nWho should attend\nResearchers and practitioners across machine learning, healthcare AI, causal inference, biostatistics, clinical development, epidemiology, AI for science, uncertainty and trustworthy ML, reasoning systems, digital health, pharma / biotech, real-world evidence, regulatory science, and translational medicine — including academia, industry, clinical research organizations, and interdisciplinary teams.\n\nTopics of Interest — Where simulation, causality, and clinical evidence meet\nWe invite work that advances reliable, intervention-aware patient world models — including benchmarks, negative results, position papers, and emerging directions.\n- Patient world models using longitudinal, multimodal clinical data\n- Clinical trial simulation, virtual trial arms, synthetic controls, and external control cohorts\n- Counterfactual outcome prediction, treatment effect modelling, and target trial emulation\n- Foundation models for EHR, clinical trials, real-world evidence, and patient timelines\n- Representation of interventions, treatment regimes, endpoints, eligibility criteria, pathways, and mechanisms of action\n- Causal representation learning, causal inference, and off-policy evaluation for intervention-aware simulation\n- Uncertainty quantification, calibration, abstention, ambiguity, and selective prediction\n- Temporal reasoning, logical consistency, clinical plausibility, and protocol-aware reasoning\n- Agentic systems for trial design, protocol interpretation, evidence synthesis, and clinical research workflows\n- Benchmarking, robustness, and validation against real-world evidence, historical trials, clinical knowledge, and downstream clinical utility\n\nCall for Papers\nWorld Models for High-Stakes Health (NeurIPS 2026) invites submissions on architectures, algorithms, theory, empirical studies, benchmarks, demonstrations, and position papers related to patient world models, intervention-aware modelling, clinical trial simulation, virtual populations, and the evaluation and reliable deployment of AI in high-stakes healthcare. Submissions must present original, unpublished work that has not appeared at NeurIPS or other archival machine-learning venues.\n\nKey dates\n- Submission deadline: September 1, 2026, AoE.\n- Notification: on or before September 29, 2026, AoE.\n- Workshop date: December 11–12, 2026 (Atlanta).\nAll deadlines follow the Anywhere on Earth (AoE) timezone.\n\nSubmission site\nSubmissions are managed via OpenReview and remain private during review. All authors should maintain up-to-date OpenReview profiles for conflict-of-interest management and paper matching.\n\nScope\nWe welcome contributions across the topics above. Accepted papers are presented as posters, with a subset selected for oral, spotlight, or demonstration talks, and we give a Best Paper Award and a Best Clinical Impact Paper Award for work with particularly strong potential to improve clinical research, healthcare delivery, or patient outcomes. The workshop is in person at NeurIPS 2026 in Atlanta.\n\nSubmission guidelines\nFormatting: Submissions must be in English and use the NeurIPS 2026 workshop LaTeX template. Papers are submitted as a single PDF:\n- Full Papers: at most 9 pages of main text.\n- Extended Abstracts: at most 4 pages of main text.\n- Demo Track: working demonstrations of systems and tools for patient modelling, clinical trial simulation, healthcare AI evaluation, or related applications, presented alongside the poster sessions.\n- Position Papers: on validation, governance, regulation, evaluation standards, and the responsible clinical use of AI.\nReferences and appendices do not count toward the page limit, but the main text must be self-contained.\n\nResponsible-use statement: Every submission also includes a short responsible-use statement covering relevant limitations, uncertainty, potential clinical or societal impacts, and suggested mitigations. It is reviewed with the paper, and a missing statement is grounds for desk rejection.\n\nAnonymity: The workshop uses double-blind review. Submissions must be anonymized, with author names, affiliations, and acknowledgments removed and prior work cited in the third person.\n\nNon-archival policy: The workshop is non-archival. Papers under review elsewhere are welcome, and accepted papers may be published at other venues afterward. Work already published at NeurIPS or other archival machine-learning venues should not be submitted.\n\nContact: Email jay.nanavati@iqvia.com.", "cfp_status": "published", "topics": [ "Patient world models using longitudinal, multimodal clinical data", "Clinical trial simulation, virtual trial arms, synthetic controls, and external control cohorts", "Counterfactual outcome prediction, treatment effect modelling, and target trial emulation", "Foundation models for EHR, clinical trials, real-world evidence, and patient timelines", "Representation of interventions, treatment regimes, endpoints, eligibility criteria, pathways, and mechanisms of action", "Causal representation learning, causal inference, and off-policy evaluation", "Uncertainty quantification, calibration, abstention, and selective prediction", "Temporal reasoning, logical consistency, clinical plausibility, and protocol-aware reasoning", "Agentic systems for trial design, protocol interpretation, and evidence synthesis", "Benchmarking, robustness, and validation against real-world evidence and downstream clinical utility" ], "important_dates": [ { "label": "Submission Deadline", "date": "2026-09-01" }, { "label": "Notification", "date": "2026-09-29" }, { "label": "Workshop", "date": "2026-12-11 or 2026-12-12" } ], "organizers": [ "Jay Nanavati (IQVIA)", "Rahul G. Krishnan (University of Toronto / Vector Institute)", "Shalmali Joshi (Columbia University)", "Lin Li (University of Oxford)", "Katie Link (NVIDIA)", "Emma Slade (GSK)" ], "speakers": [ "Nathan Kallus (Cornell University)", "Chris Tomlinson (UCL / NHS Foresight)", "Raja Shankar (IQVIA) — panel moderator", "Rose Yu (UC San Diego) — panelist" ], "host_url": "https://wmhs-neurips.github.io/WMHS/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/WMHS", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/WMHS", "location": "Atlanta, United States", "city": "Atlanta", "workshop_date": "2026-12-12", "contact": "jay.nanavati@iqvia.com", "tracks": [ { "key": "WMHS", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/WMHS", "submission_dates_raw": "" } ], "group": "health", "group_label": "Health & Medicine", "corpus": "World Models for High-Stakes Health: Reliable Clinical Trial Simulation and Intervention-Aware Reasoning @ NeurIPS 2026 WMHS 2026 The first WMHS workshop focuses on patient world models — generative and reasoning-capable systems that learn from patient trajectories, trial protocols, interventions, and real-world evidence — using clinical trial simulation as a demanding, falsifiable testbed for building reliable, intervention-aware world models in high-stakes healthcare. Patient world models using longitudinal, multimodal clinical data Clinical trial simulation, virtual trial arms, synthetic controls, and external control cohorts Counterfactual outcome prediction, treatment effect modelling, and target trial emulation Foundation models for EHR, clinical trials, real-world evidence, and patient timelines Representation of interventions, treatment regimes, endpoints, eligibility criteria, pathways, and mechanisms of action Causal representation learning, causal inference, and off-policy evaluation Uncertainty quantification, calibration, abstention, and selective prediction Temporal reasoning, logical consistency, clinical plausibility, and protocol-aware reasoning Agentic systems for trial design, protocol interpretation, and evidence synthesis Benchmarking, robustness, and validation against real-world evidence and downstream clinical utility World Models for High-Stakes Health: Reliable Clinical Trial Simulation and Intervention-Aware Reasoning\n\nAbout the workshop — Clinical trial simulation as a stress test for world models\nClinical trial simulation is a uniquely demanding and falsifiable testbed for one of the central open problems in machine learning: building reliable world models of complex, partially observed, intervention-rich systems.\n\nThis workshop focuses on patient world models — generative and reasoning-capable systems that learn from patient trajectories, trial protocols, interventions, mechanisms of action, and real-world evidence to support clinical trial simulation and decision-making.\n\nWhy now\nFoundation models, sequence models, and generative simulators are converging with healthcare AI's shift from static risk prediction toward longitudinal modelling, treatment response estimation, and real-world evidence — yet reliability in high-stakes medical settings remains unresolved.\n\nWho should attend\nResearchers and practitioners across machine learning, healthcare AI, causal inference, biostatistics, clinical development, epidemiology, AI for science, uncertainty and trustworthy ML, reasoning systems, digital health, pharma / biotech, real-world evidence, regulatory science, and translational medicine — including academia, industry, clinical research organizations, and interdisciplinary teams.\n\nTopics of Interest — Where simulation, causality, and clinical evidence meet\nWe invite work that advances reliable, intervention-aware patient world models — including benchmarks, negative results, position papers, and emerging directions.\n- Patient world models using longitudinal, multimodal clinical data\n- Clinical trial simulation, virtual trial arms, synthetic controls, and external control cohorts\n- Counterfactual outcome prediction, treatment effect modelling, and target trial emulation\n- Foundation models for EHR, clinical trials, real-world evidence, and patient timelines\n- Representation of interventions, treatment regimes, endpoints, eligibility criteria, pathways, and mechanisms of action\n- Causal representation learning, causal inference, and off-policy evaluation for intervention-aware simulation\n- Uncertainty quantification, calibration, abstention, ambiguity, and selective prediction\n- Temporal reasoning, logical consistency, clinical plausibility, and protocol-aware reasoning\n- Agentic systems for trial design, protocol interpretation, evidence synthesis, and clinical research workflows\n- Benchmarking, robustness, and validation against real-world evidence, historical trials, clinical knowledge, and downstream clinical utility\n\nCall for Papers\nWorld Models for High-Stakes Health (NeurIPS 2026) invites submissions on architectures, algorithms, theory, empirical studies, benchmarks, demonstrations, and position papers related to patient world models, intervention-aware modelling, clinical trial simulation, virtual populations, and the evaluation and reliable deployment of AI in high-stakes healthcare. Submissions must present original, unpublished work that has not appeared at NeurIPS or other archival machine-learning venues.\n\nKey dates\n- Submission deadline: September 1, 2026, AoE.\n- Notification: on or before September 29, 2026, AoE.\n- Workshop date: December 11–12, 2026 (Atlanta).\nAll deadlines follow the Anywhere on Earth (AoE) timezone.\n\nSubmission site\nSubmissions are managed via OpenReview and remain private during review. All authors should maintain up-to-date OpenReview profiles for conflict-of-interest management and paper matching.\n\nScope\nWe welcome contributions across the topics above. Accepted papers are presented as posters, with a subset selected for oral, spotlight, or demonstration talks, and we give a Best Paper Award and a Best Clinical Impact Paper Award for work with particularly strong potential to improve clinical research, healthcare delivery, or patient outcomes. The workshop is in person at NeurIPS 2026 in Atlanta.\n\nSubmission guidelines\nFormatting: Submissions must be in English and use the NeurIPS 2026 workshop LaTeX template. Papers are submitted as a single PDF:\n- Full Papers: at most 9 pages of main text.\n- Extended Abstracts: at most 4 pages of main text.\n- Demo Track: working demonstrations of systems and tools for patient modelling, clinical trial simulation, healthcare AI evaluation, or related applications, presented alongside the poster sessions.\n- Position Papers: on validation, governance, regulation, evaluation standards, and the responsible clinical use of AI.\nReferences and appendices do not count toward the page limit, but the main text must be self-contained.\n\nResponsible-use statement: Every submission also includes a short responsible-use statement covering relevant limitations, uncertainty, potential clinical or societal impacts, and suggested mitigations. It is reviewed with the paper, and a missing statement is grounds for desk rejection.\n\nAnonymity: The workshop uses double-blind review. Submissions must be anonymized, with author names, affiliations, and acknowledgments removed and prior work cited in the third person.\n\nNon-archival policy: The workshop is non-archival. Papers under review elsewhere are welcome, and accepted papers may be published at other venues afterward. Work already published at NeurIPS or other archival machine-learning venues should not be submitted.\n\nContact: Email jay.nanavati@iqvia.com." }, { "key": "WM_PAI", "title": "World Models in Physical AI Workshop", "subtitle": "WM PAI Workshop", "summary": "A one-day NeurIPS 2026 workshop on learned models of physical-world dynamics for Physical AI, bringing together generative modeling, reinforcement learning, robotics, computer vision, autonomous driving, and simulation to make world models actionable, physically grounded, and deployable.", "cfp_full": "World Models in Physical AI\n\nLearned models that simulate, plan, and act in the real world — bringing together generative modeling, reinforcement learning, robotics, computer vision, autonomous driving, and simulation.\n\nAbout the workshop\n\nMaking world models actionable, physically grounded, and deployable\n\nWorld models predict how an environment evolves, optionally conditioned on an agent's actions, and can be rolled out to imagine future observations and outcomes. For physical-AI systems — robots, autonomous vehicles, and embodied agents — this capability is central: an agent that can simulate its world can plan, learn from imagined experience, and handle situations missing from its training data.\n\nProgress now spans latent-dynamics models for control, interactive and generative video simulators, foundation world-model platforms, and closed-loop autonomous-driving simulation. Model-based RL, generative simulation, and video prediction are all converging on controllable models of the world — but in separate communities. This one-day workshop brings them together to ask what world models must deliver to serve as a deployable computational substrate for physical AI.\n\nScope — Topics we invite\n\nTalks, papers, and discussion spanning the pipeline from data and representations, through evaluation, to planning and control.\n\n- Representations & architectures: Latent vs. pixel/video models, JEPA embeddings, identifiable latents, omnimodal models, 3D and contact dynamics.\n- World models for action: Model-based RL, planning and control, learning in imagination, forward and inverse dynamics, world-action models.\n- Generative simulation: World models as data generators and interactive closed-loop simulators for robotics and driving; sim-to-real and real-to-sim.\n- Evaluation: Measuring whether world models are physically correct, causally faithful, and useful for downstream control; benchmarking robustness and generalization.\n- Scaling & foundation models: Data, compute, and generalization of large pretrained world models across embodiments and domains.\n- Safety & broader impact: Reliability, safety, and the broader impact of world models that drive real physical systems.\n\nThe debate — Open questions we'll take positions on\n\n1. What does \"correct\" mean? Can we agree on benchmarks that measure physical consistency and downstream control utility — not just pixel fidelity? For closed-loop simulators, should evaluation preserve policy rankings and real-world failure modes?\n2. Pixels vs. latents. Should world models predict in observation space or in abstract latent space? When are inspectable rollouts necessary, and when are identifiable, low-dimensional latents enough for planning and control?\n3. Controllability & long horizons. How do we keep rollouts action-controllable and physically consistent over long horizons — preserving object identity, contact, causality, and state under repeated interventions?\n4. Sim-to-real & real-to-sim. When can a generative world model replace or augment a physics or reconstruction-based simulator for training and evaluating deployable policies?\n5. Scaling laws. Do world models for physical AI follow favorable scaling laws, and what multimodal data unlocks generalization to new embodiments, action spaces, and domains?\n\nCall for papers\n\nWe invite submissions on learned models of physical-world dynamics for Physical AI. Papers may be up to 8 pages, excluding references, using the NeurIPS 2026 template. Accepted work is non-archival and managed on OpenReview.\n\n- Up to 8 pages, excluding references, using the NeurIPS 2026 paper template.\n- Non-archival, on OpenReview. Authors may later submit to an archival venue. At least one author per submission must agree to review.\n- Reviewing is double-blind; anonymization is required.\n- AI tool policy: Authors may use assistance tools but remain fully responsible for content accuracy. If agents or LLMs are an important, original or non-standard component of the research method, their use must be described.\n- Accepted papers present via posters; selected works receive spotlight talks.\n\nImportant dates (all AoE):\n- Late July 2026: Call for papers released\n- August 29, 2026: Submission deadline\n- Aug 30 – Sept 20, 2026: Reviewing period\n- September 29, 2026: Author notification\n- November 2026: Camera-ready & final posters\n- December 12 or 13, 2026: Workshop day · Sydney\n\nCo-located challenge: AV Causal Reasoning Retrieval Challenge — a challenge on causal reasoning and retrieval for autonomous driving, run alongside the workshop. Winners will be announced during the workshop.", "cfp_status": "published", "topics": [ "Representations & architectures (latent vs. pixel/video models, JEPA embeddings, identifiable latents, omnimodal models, 3D and contact dynamics)", "World models for action (model-based RL, planning and control, learning in imagination, forward and inverse dynamics, world-action models)", "Generative simulation (data generators, interactive closed-loop simulators for robotics and driving, sim-to-real and real-to-sim)", "Evaluation (physical correctness, causal faithfulness, benchmarking robustness and generalization)", "Scaling & foundation models (data, compute, and generalization across embodiments and domains)", "Safety & broader impact of world models driving real physical systems" ], "important_dates": [ { "label": "Call for papers released", "date": "Late July 2026" }, { "label": "Submission deadline", "date": "2026-08-29" }, { "label": "Reviewing period", "date": "2026-08-30 to 2026-09-20" }, { "label": "Author notification", "date": "2026-09-29" }, { "label": "Camera-ready & final posters", "date": "November 2026" }, { "label": "Workshop day", "date": "2026-12-12" } ], "organizers": [ "Jenny Schmalfuss (Research Scientist, NVIDIA)", "German Ros (Principal Scientist, NVIDIA)", "Despoina Paschalidou (Research Scientist, NVIDIA)", "Roberto Martín-Martín (Assistant Professor, UT Austin)", "Jose M. Alvarez (Director of Research, NVIDIA)" ], "speakers": [ "Mengyue Yang (Assistant Professor, University of Bristol)", "Katerina Fragkiadaki (Associate Professor, Carnegie Mellon University)", "Jan Eric Lenssen (Senior Researcher, MPI for Informatics)", "Max Jiang (Staff Research Scientist, Waymo)", "Danijar Hafner (Staff Research Scientist, Google DeepMind)" ], "host_url": "http://www.worldmodels-physicalai.com/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/WM_PAI", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/WM_PAI", "location": "Sydney, Australia", "city": "Sydney", "workshop_date": "2026-12-11", "contact": "grossanchez@nvidia.com", "tracks": [ { "key": "WM_PAI", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/WM_PAI", "submission_dates_raw": "" } ], "group": "agents", "group_label": "Agents & Agentic Systems", "corpus": "World Models in Physical AI Workshop WM PAI Workshop A one-day NeurIPS 2026 workshop on learned models of physical-world dynamics for Physical AI, bringing together generative modeling, reinforcement learning, robotics, computer vision, autonomous driving, and simulation to make world models actionable, physically grounded, and deployable. Representations & architectures (latent vs. pixel/video models, JEPA embeddings, identifiable latents, omnimodal models, 3D and contact dynamics) World models for action (model-based RL, planning and control, learning in imagination, forward and inverse dynamics, world-action models) Generative simulation (data generators, interactive closed-loop simulators for robotics and driving, sim-to-real and real-to-sim) Evaluation (physical correctness, causal faithfulness, benchmarking robustness and generalization) Scaling & foundation models (data, compute, and generalization across embodiments and domains) Safety & broader impact of world models driving real physical systems World Models in Physical AI\n\nLearned models that simulate, plan, and act in the real world — bringing together generative modeling, reinforcement learning, robotics, computer vision, autonomous driving, and simulation.\n\nAbout the workshop\n\nMaking world models actionable, physically grounded, and deployable\n\nWorld models predict how an environment evolves, optionally conditioned on an agent's actions, and can be rolled out to imagine future observations and outcomes. For physical-AI systems — robots, autonomous vehicles, and embodied agents — this capability is central: an agent that can simulate its world can plan, learn from imagined experience, and handle situations missing from its training data.\n\nProgress now spans latent-dynamics models for control, interactive and generative video simulators, foundation world-model platforms, and closed-loop autonomous-driving simulation. Model-based RL, generative simulation, and video prediction are all converging on controllable models of the world — but in separate communities. This one-day workshop brings them together to ask what world models must deliver to serve as a deployable computational substrate for physical AI.\n\nScope — Topics we invite\n\nTalks, papers, and discussion spanning the pipeline from data and representations, through evaluation, to planning and control.\n\n- Representations & architectures: Latent vs. pixel/video models, JEPA embeddings, identifiable latents, omnimodal models, 3D and contact dynamics.\n- World models for action: Model-based RL, planning and control, learning in imagination, forward and inverse dynamics, world-action models.\n- Generative simulation: World models as data generators and interactive closed-loop simulators for robotics and driving; sim-to-real and real-to-sim.\n- Evaluation: Measuring whether world models are physically correct, causally faithful, and useful for downstream control; benchmarking robustness and generalization.\n- Scaling & foundation models: Data, compute, and generalization of large pretrained world models across embodiments and domains.\n- Safety & broader impact: Reliability, safety, and the broader impact of world models that drive real physical systems.\n\nThe debate — Open questions we'll take positions on\n\n1. What does \"correct\" mean? Can we agree on benchmarks that measure physical consistency and downstream control utility — not just pixel fidelity? For closed-loop simulators, should evaluation preserve policy rankings and real-world failure modes?\n2. Pixels vs. latents. Should world models predict in observation space or in abstract latent space? When are inspectable rollouts necessary, and when are identifiable, low-dimensional latents enough for planning and control?\n3. Controllability & long horizons. How do we keep rollouts action-controllable and physically consistent over long horizons — preserving object identity, contact, causality, and state under repeated interventions?\n4. Sim-to-real & real-to-sim. When can a generative world model replace or augment a physics or reconstruction-based simulator for training and evaluating deployable policies?\n5. Scaling laws. Do world models for physical AI follow favorable scaling laws, and what multimodal data unlocks generalization to new embodiments, action spaces, and domains?\n\nCall for papers\n\nWe invite submissions on learned models of physical-world dynamics for Physical AI. Papers may be up to 8 pages, excluding references, using the NeurIPS 2026 template. Accepted work is non-archival and managed on OpenReview.\n\n- Up to 8 pages, excluding references, using the NeurIPS 2026 paper template.\n- Non-archival, on OpenReview. Authors may later submit to an archival venue. At least one author per submission must agree to review.\n- Reviewing is double-blind; anonymization is required.\n- AI tool policy: Authors may use assistance tools but remain fully responsible for content accuracy. If agents or LLMs are an important, original or non-standard component of the research method, their use must be described.\n- Accepted papers present via posters; selected works receive spotlight talks.\n\nImportant dates (all AoE):\n- Late July 2026: Call for papers released\n- August 29, 2026: Submission deadline\n- Aug 30 – Sept 20, 2026: Reviewing period\n- September 29, 2026: Author notification\n- November 2026: Camera-ready & final posters\n- December 12 or 13, 2026: Workshop day · Sydney\n\nCo-located challenge: AV Causal Reasoning Retrieval Challenge — a challenge on causal reasoning and retrieval for autonomous driving, run alongside the workshop. Winners will be announced during the workshop." }, { "key": "XAI4Science", "title": "XAI4Science: Knowledge Discovery and Trust through Interpretable Foundation Models", "subtitle": "XAI4Science", "summary": "A workshop at the intersection of explainable AI, foundation models, and climate science, bringing together ML researchers, climate scientists, and policy experts to build trustworthy, interpretable foundation models for weather and climate forecasting.", "cfp_full": "About\n\nEfficient and Interpretable Foundation Models\n\nAccurate weather and climate forecasting models enable researchers and policymakers to make informed decisions towards a more sustainable future. However, traditional numerical models demand immense computational resources. Machine Learning (ML) has emerged as a viable and efficient alternative, benefiting from the ever-growing volume of collected data. In particular, weather and climate Foundation Models (FMs) achieve comparable results at a fraction of the computational cost, opening the field to smaller countries and research teams with limited resources and representation. Yet while FMs partly address computational costs issues, they introduce a transparency one. Unlike physics-based models, whose predictions can be traced to known equations with a relatively well understood physics, FMs results may prove difficult to understand, especially when they produce physical hallucinations. This opacity may undermine scientific trust and raise equity concerns.\n\nExplainable Artificial Intelligence (XAI) offers a promising path to address this challenge. By revealing which inputs and internal mechanisms drive a model's predictions, XAI methods can help verify that FMs rely on physically meaningful patterns rather than spurious correlations, detect and diagnose physical hallucinations, and communicate the rationale behind forecasts to scientists and decision makers alike. In doing so, XAI can restore the scientific scrutiny that black-box models currently lack, fostering trust in ML-based forecasting and supporting its equitable adoption across the globe. In this workshop we focus on the issues just introduced, by bringing together ML researchers, climate scientists, and policy experts.\n\nWhat the workshop covers\n\nBy bringing together ML researchers, climate scientists, and policy experts, the program focuses on three threads:\n- Ante-hoc interpretability: Self-explainable architectures and inductive biases for weather and climate Foundation Models.\n- Post-hoc attribution & evaluation: Attribution, probing, and mechanistic interpretability methods, including rigorous evaluation of their faithfulness.\n- Physics-consistent explanations: Benchmarks and methods for validating explanations against known physical laws and causal structure.\n\nCall for Submissions\n\nSubmission Tracks\n(1) Regular Paper Track: up to 8 pages, excluding references and appendices.\n(2) Tiny Paper Track: up to 5 pages, excluding references and appendices.\nUnlimited pages are allowed for references and appendices in the same PDF as the main paper.\n\nSubmission Format\nSubmissions must be in a single PDF file and are required to use the NeurIPS 2026 LaTeX template, available on the NeurIPS 2026 Main Track format. No Paper Checklist is required for workshop submissions.\n\nSubmission Link\nAll submissions must be made via OpenReview. Please make sure that all authors have an OpenReview profile with the latest information. Creating one may take up to 2 weeks.\n\nGeneral Policy\nThe workshop is non-archival and does not publish proceedings. Submissions can be subsequently or concurrently submitted to other venues. We welcome optional anonymous submissions of ongoing and unpublished work on any topics related to the workshop. We require each submission to nominate at least one author to serve as a reviewer, following the NeurIPS 2026 reciprocal review rule, and we aim for each paper to collect at least three reviews.\n\nContributed Talks\nContributed talks' speakers will be selected from the top submissions received, giving deserving works a spotlight to a broad audience. Each speaker will have 15 minutes (10+5) for presentation and Q&A.\n\nImportant Dates\nAll dates are in AoE (Anywhere on Earth, UTC-12) time. Deadlines end at 23:59 AoE, equivalent to 11:59 GMT the following day.\nSubmission Deadline: August 29, 2026, AoE\nReview Period: September 1 - September 15, 2026, AoE\nRebuttal Period Ends: September 20, 2026, AoE\nAdvisory Committee Discussion Ends: September 27, 2026, AoE\nNotification of Acceptance: September 28, 2026, AoE\nWorkshop Date: December 11 or 12, 2026, NeurIPS 2026", "cfp_status": "published", "topics": [ "Ante-hoc interpretability: self-explainable architectures and inductive biases for weather and climate Foundation Models", "Post-hoc attribution & evaluation: attribution, probing, and mechanistic interpretability methods, including rigorous evaluation of their faithfulness", "Physics-consistent explanations: benchmarks and methods for validating explanations against known physical laws and causal structure", "Explainable AI for weather and climate foundation models", "Detecting and diagnosing physical hallucinations in foundation models", "Trust and equity in ML-based forecasting" ], "important_dates": [ { "label": "Submission Start", "date": "2026-08-01" }, { "label": "Submission Deadline", "date": "2026-08-29" }, { "label": "Review Period", "date": "2026-09-01" }, { "label": "Rebuttal Period Ends", "date": "2026-09-20" }, { "label": "Advisory Committee Discussion Ends", "date": "2026-09-27" }, { "label": "Notification of Acceptance", "date": "2026-09-28" }, { "label": "Workshop Date", "date": "December 11 or 12, 2026" } ], "organizers": [ "Leonardo Pesce (National University of Singapore)", "Jiawen Wei (National University of Singapore)", "Max Welling (University of Amsterdam)", "Gianmarco Mengaldo (National University of Singapore)" ], "speakers": [ "Gustau Camps-Valls (Universitat de Valencia)", "Megan J. Stanley (Ellison Institute of Technology Oxford)", "Lily Xu (Columbia University)", "David Rolnick (McGill University & Mila)" ], "host_url": "https://xai4science.github.io/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/XAI4Science", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/XAI4Science", "location": "Sydney, Australia", "city": "Sydney", "workshop_date": "", "contact": "xai4science@gmail.com", "tracks": [ { "key": "XAI4Science", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/XAI4Science", "submission_dates_raw": "Submission Start: Aug 01 2026 12:00PM UTC-0, Submission Deadline: Aug 30 2026 12:00PM UTC-0" } ], "group": "trust", "group_label": "Trustworthy & Safe AI", "corpus": "XAI4Science: Knowledge Discovery and Trust through Interpretable Foundation Models XAI4Science A workshop at the intersection of explainable AI, foundation models, and climate science, bringing together ML researchers, climate scientists, and policy experts to build trustworthy, interpretable foundation models for weather and climate forecasting. Ante-hoc interpretability: self-explainable architectures and inductive biases for weather and climate Foundation Models Post-hoc attribution & evaluation: attribution, probing, and mechanistic interpretability methods, including rigorous evaluation of their faithfulness Physics-consistent explanations: benchmarks and methods for validating explanations against known physical laws and causal structure Explainable AI for weather and climate foundation models Detecting and diagnosing physical hallucinations in foundation models Trust and equity in ML-based forecasting About\n\nEfficient and Interpretable Foundation Models\n\nAccurate weather and climate forecasting models enable researchers and policymakers to make informed decisions towards a more sustainable future. However, traditional numerical models demand immense computational resources. Machine Learning (ML) has emerged as a viable and efficient alternative, benefiting from the ever-growing volume of collected data. In particular, weather and climate Foundation Models (FMs) achieve comparable results at a fraction of the computational cost, opening the field to smaller countries and research teams with limited resources and representation. Yet while FMs partly address computational costs issues, they introduce a transparency one. Unlike physics-based models, whose predictions can be traced to known equations with a relatively well understood physics, FMs results may prove difficult to understand, especially when they produce physical hallucinations. This opacity may undermine scientific trust and raise equity concerns.\n\nExplainable Artificial Intelligence (XAI) offers a promising path to address this challenge. By revealing which inputs and internal mechanisms drive a model's predictions, XAI methods can help verify that FMs rely on physically meaningful patterns rather than spurious correlations, detect and diagnose physical hallucinations, and communicate the rationale behind forecasts to scientists and decision makers alike. In doing so, XAI can restore the scientific scrutiny that black-box models currently lack, fostering trust in ML-based forecasting and supporting its equitable adoption across the globe. In this workshop we focus on the issues just introduced, by bringing together ML researchers, climate scientists, and policy experts.\n\nWhat the workshop covers\n\nBy bringing together ML researchers, climate scientists, and policy experts, the program focuses on three threads:\n- Ante-hoc interpretability: Self-explainable architectures and inductive biases for weather and climate Foundation Models.\n- Post-hoc attribution & evaluation: Attribution, probing, and mechanistic interpretability methods, including rigorous evaluation of their faithfulness.\n- Physics-consistent explanations: Benchmarks and methods for validating explanations against known physical laws and causal structure.\n\nCall for Submissions\n\nSubmission Tracks\n(1) Regular Paper Track: up to 8 pages, excluding references and appendices.\n(2) Tiny Paper Track: up to 5 pages, excluding references and appendices.\nUnlimited pages are allowed for references and appendices in the same PDF as the main paper.\n\nSubmission Format\nSubmissions must be in a single PDF file and are required to use the NeurIPS 2026 LaTeX template, available on the NeurIPS 2026 Main Track format. No Paper Checklist is required for workshop submissions.\n\nSubmission Link\nAll submissions must be made via OpenReview. Please make sure that all authors have an OpenReview profile with the latest information. Creating one may take up to 2 weeks.\n\nGeneral Policy\nThe workshop is non-archival and does not publish proceedings. Submissions can be subsequently or concurrently submitted to other venues. We welcome optional anonymous submissions of ongoing and unpublished work on any topics related to the workshop. We require each submission to nominate at least one author to serve as a reviewer, following the NeurIPS 2026 reciprocal review rule, and we aim for each paper to collect at least three reviews.\n\nContributed Talks\nContributed talks' speakers will be selected from the top submissions received, giving deserving works a spotlight to a broad audience. Each speaker will have 15 minutes (10+5) for presentation and Q&A.\n\nImportant Dates\nAll dates are in AoE (Anywhere on Earth, UTC-12) time. Deadlines end at 23:59 AoE, equivalent to 11:59 GMT the following day.\nSubmission Deadline: August 29, 2026, AoE\nReview Period: September 1 - September 15, 2026, AoE\nRebuttal Period Ends: September 20, 2026, AoE\nAdvisory Committee Discussion Ends: September 27, 2026, AoE\nNotification of Acceptance: September 28, 2026, AoE\nWorkshop Date: December 11 or 12, 2026, NeurIPS 2026" }, { "key": "GlobalSouthAI", "title": "NeurIPS 2026 GlobalSouthAI: Rethinking AI for and from the Global South", "subtitle": "NeurIPS 2026 GlobalSouthAI Affinity Event", "summary": "GlobalSouthAI is a NeurIPS 2026 affinity event focused on rethinking AI for and from the Global South, featuring keynote talks, a 3-Minute Presentation (3MT) contest, and a poster session across Sydney and Paris.", "cfp_full": "Global South AI - Rethinking AI for and from the Global South\n\nNeurIPS 2026 Affinity Event | December 2026 | 8:00 AM - 5:00 PM | Sydney, Australia and Paris, France (TBD)\n\nThis is an affinity event focused on \"Rethinking AI for and from the Global South.\"\n\nRegistration Details: To be updated.\n\nProgram Schedule: TBD\n\nAccepted Works:\n- 3-Minute Presentation (3MT) Contest Finalists: TBD\n- Poster Session Paper IDs: TBD\n\nContact: globalsouthai-neurips-26@googlegroups.com", "cfp_status": "not_yet_published", "topics": [ "AI for and from the Global South" ], "important_dates": [], "organizers": [ "Tushar Shinde (Chair, IIT Madras Zanzibar, Tanzania)", "Saurabh Deshpande (Birla AI Labs, India)", "Sandareka Wickramanayake (University of Moratuwa, Sri Lanka)", "Kumar Rahul (Amazon Prime Video, USA)", "Akila Hewa Thondilege (Queensland University of Technology, Australia)", "Jonathan Shock (University of Cape Town, South Africa)", "Divya Sharma (York University, Canada)", "Chaker Larabi (Universite de Poitiers, France)" ], "speakers": [ "Patrick Le Callet (Nantes Universite, France)", "Sunayana Sitaram (Microsoft Research, India)", "Maria G. Martini (Kingston University London, UK)", "Monojit Choudhury (MBZUAI, UAE)", "Balaraman Ravindran (IIT Madras, India)" ], "host_url": "https://sites.google.com/view/globalsouthai-neurips26/home", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/GlobalSouthAI", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/GlobalSouthAI", "location": "Sydney, Australia and Paris, France", "city": "Multiple", "workshop_date": "2026-12-05", "contact": "shinde@iitmz.ac.in", "tracks": [ { "key": "GlobalSouthAI", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/GlobalSouthAI", "submission_dates_raw": "" } ], "group": "society", "group_label": "Society, Applications & Community", "corpus": "NeurIPS 2026 GlobalSouthAI: Rethinking AI for and from the Global South NeurIPS 2026 GlobalSouthAI Affinity Event GlobalSouthAI is a NeurIPS 2026 affinity event focused on rethinking AI for and from the Global South, featuring keynote talks, a 3-Minute Presentation (3MT) contest, and a poster session across Sydney and Paris. AI for and from the Global South Global South AI - Rethinking AI for and from the Global South\n\nNeurIPS 2026 Affinity Event | December 2026 | 8:00 AM - 5:00 PM | Sydney, Australia and Paris, France (TBD)\n\nThis is an affinity event focused on \"Rethinking AI for and from the Global South.\"\n\nRegistration Details: To be updated.\n\nProgram Schedule: TBD\n\nAccepted Works:\n- 3-Minute Presentation (3MT) Contest Finalists: TBD\n- Poster Session Paper IDs: TBD\n\nContact: globalsouthai-neurips-26@googlegroups.com" }, { "key": "EvoRobust", "title": "NeurIPS 2026 Workshop - Self-Evolving Diversity-Driven Search for Robust AI Systems", "subtitle": "EvoRobust@NeurIPS 2026", "summary": "A NeurIPS 2026 workshop on self-evolving, diversity-driven search methods for building robust AI systems.", "cfp_full": "", "cfp_status": "not_yet_published", "topics": [], "important_dates": [ { "label": "Submission Deadline", "date": "2026-08-30" } ], "organizers": [], "speakers": [], "host_url": "https://neurips.cc/Conferences/2026", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/EvoRobust", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/EvoRobust", "location": "Sydney, Australia", "city": "Sydney", "workshop_date": "", "contact": "evorobust-workshop@googlegroups.com", "tracks": [ { "key": "EvoRobust", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/EvoRobust", "submission_dates_raw": "Submission Deadline: Aug 30 2026 12:29PM UTC-0" } ], "group": "theory", "group_label": "Theory & Methods", "corpus": "NeurIPS 2026 Workshop - Self-Evolving Diversity-Driven Search for Robust AI Systems EvoRobust@NeurIPS 2026 A NeurIPS 2026 workshop on self-evolving, diversity-driven search methods for building robust AI systems. " }, { "key": "BrainBodyFM", "title": "NeurIPS 2026 Workshop on Foundation Models for the Brain and Body", "subtitle": "NeurIPS 2026 Workshop BrainBodyFM", "summary": "A workshop bringing together neuroscientists, biomedical engineers, wearable technology researchers, and machine learning experts advancing foundation model approaches for neural and physiological biosignals (EEG, intracortical electrophysiology, EMG, MEG, ECG), toward AI models that capture the complexity of the brain, body, and behavior at scale.", "cfp_full": "", "cfp_status": "not_yet_published", "topics": [ "Foundation models for brain and physiological biosignals", "EEG, intracortical electrophysiology, EMG, MEG, and ECG modeling", "Brain-computer interfacing", "Wearable and health monitoring signals", "Learning from noisy, heterogeneous biosignal timeseries across subjects, devices, and environments" ], "important_dates": [], "organizers": [], "speakers": [], "host_url": "https://brainbodyfm-workshop.github.io", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/BrainBodyFM", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/BrainBodyFM", "location": "Sydney, Australia", "city": "Sydney", "workshop_date": "2026-12-10", "contact": "brainbodyfm@googlegroups.com", "tracks": [ { "key": "BrainBodyFM", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/BrainBodyFM", "submission_dates_raw": "" } ], "group": "health", "group_label": "Health & Medicine", "corpus": "NeurIPS 2026 Workshop on Foundation Models for the Brain and Body NeurIPS 2026 Workshop BrainBodyFM A workshop bringing together neuroscientists, biomedical engineers, wearable technology researchers, and machine learning experts advancing foundation model approaches for neural and physiological biosignals (EEG, intracortical electrophysiology, EMG, MEG, ECG), toward AI models that capture the complexity of the brain, body, and behavior at scale. Foundation models for brain and physiological biosignals EEG, intracortical electrophysiology, EMG, MEG, and ECG modeling Brain-computer interfacing Wearable and health monitoring signals Learning from noisy, heterogeneous biosignal timeseries across subjects, devices, and environments " }, { "key": "GEM_Bio", "title": "NeurIPS Workshop on Integrating Generative and Experimental Platforms for Biomolecular Design (GEM)", "subtitle": "NeurIPS 2026 Workshop GEM Bio", "summary": "The GEM (Generative and Experimental Perspectives for Biomolecular Design) workshop brings computationalists and experimentalists together to bridge generative machine learning and experimental biology for the design of proteins, molecules, and nucleic acids, featuring an in-silico (dry-lab) track and an experimental (wet-lab) track.", "cfp_full": "", "cfp_status": "not_yet_published", "topics": [ "Inverse design of biomolecules (proteins, molecules, nucleic acids)", "Modelling biomolecular data", "Model interpretability", "Generative ML with experimental (wet-lab) results on biological/chemical problems", "High-throughput data methods", "Active learning", "Benchmarks, datasets, and oracles" ], "important_dates": [], "organizers": [], "speakers": [], "host_url": "https://www.gembio.ai/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/GEM_Bio", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/GEM_Bio", "location": "Atlanta, Georgia", "city": "Atlanta", "workshop_date": "2026-08-03", "contact": "gembioworkshop@googlegroups.com", "tracks": [ { "key": "GEM_Bio", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/GEM_Bio", "submission_dates_raw": "" } ], "group": "science", "group_label": "AI for Science", "corpus": "NeurIPS Workshop on Integrating Generative and Experimental Platforms for Biomolecular Design (GEM) NeurIPS 2026 Workshop GEM Bio The GEM (Generative and Experimental Perspectives for Biomolecular Design) workshop brings computationalists and experimentalists together to bridge generative machine learning and experimental biology for the design of proteins, molecules, and nucleic acids, featuring an in-silico (dry-lab) track and an experimental (wet-lab) track. Inverse design of biomolecules (proteins, molecules, nucleic acids) Modelling biomolecular data Model interpretability Generative ML with experimental (wet-lab) results on biological/chemical problems High-throughput data methods Active learning Benchmarks, datasets, and oracles " }, { "key": "WiML", "title": "Women in Machine Learning Workshop @ NeurIPS 2026", "subtitle": "WiML @ NeurIPS 2026", "summary": "The Women in Machine Learning (WiML) Workshop is a technical affinity workshop co-located with NeurIPS 2026 that showcases the research and technical accomplishments of women and gender minorities in machine learning, providing mentorship, visibility, and community-building opportunities.", "cfp_full": "", "cfp_status": "not_yet_published", "topics": [], "important_dates": [], "organizers": [], "speakers": [], "host_url": "https://www.wiml.org/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/WiML", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/WiML", "location": "Sydney, Paris, Atlanta", "city": "Multiple", "workshop_date": "2026-12-07", "contact": "workshop@wimlworkshop.org", "tracks": [ { "key": "WiML", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/WiML", "submission_dates_raw": "" } ], "group": "society", "group_label": "Society, Applications & Community", "corpus": "Women in Machine Learning Workshop @ NeurIPS 2026 WiML @ NeurIPS 2026 The Women in Machine Learning (WiML) Workshop is a technical affinity workshop co-located with NeurIPS 2026 that showcases the research and technical accomplishments of women and gender minorities in machine learning, providing mentorship, visibility, and community-building opportunities. " }, { "key": "EconML", "title": "Workshop on Economics for Machine Learning at NeurIPS 2026", "subtitle": "EconML NeurIPS2026", "summary": "A NeurIPS 2026 workshop at the intersection of economics and machine learning, focused on strategic behavior, incentives, and mechanism/information design questions arising in learning systems.", "cfp_full": "", "cfp_status": "not_yet_published", "topics": [], "important_dates": [ { "label": "Abstract Registration", "date": "2026-09-01" }, { "label": "Submission Deadline", "date": "2026-09-07" } ], "organizers": [], "speakers": [], "host_url": "https://safwanhossain.github.io/", "submission_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/EconML", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/EconML", "location": "Atlanta, Georgia, United States", "city": "Atlanta", "workshop_date": "", "contact": "edens@caltech.edu", "tracks": [ { "key": "EconML", "openreview_url": "https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/EconML", "submission_dates_raw": "Abstract Registration: Sep 01 2026 12:00AM UTC-0, Submission Deadline: Sep 07 2026 12:00AM UTC-0" } ], "group": "society", "group_label": "Society, Applications & Community", "corpus": "Workshop on Economics for Machine Learning at NeurIPS 2026 EconML NeurIPS2026 A NeurIPS 2026 workshop at the intersection of economics and machine learning, focused on strategic behavior, incentives, and mechanism/information design questions arising in learning systems. " } ] }