Workshop on Multi-Agent Learning and Its Opportunities in the Era of Generative AI

About

The rapid emergence of generative AI has revitalized interest in multi-agent learning as a foundation for building systems that can reason, coordinate, and adapt across diverse environments. This workshop seeks to explore the growing convergence between multi-agent learning and generative AI, emphasizing their mutual potential to advance both theoretical understanding and practical capability.

We focus on three interrelated fronts where this integration is most visible:

1. LLM-based multi-agent systems, where large language models interact, cooperate, or compete in structured settings;
2. Real-world distributed system control, where multi-agent learning offers scalable and data-driven coordination strategies for complex real-world systems such as smart cities;
3. Human-AI interaction, where generative AI enables richer modelling of human preferences, values, and behaviours, supporting more human-aligned multi-agent systems.

By bringing together researchers from machine learning, game theory, cognitive science, and human-computer interaction, this workshop aims to bridge methodological insights and emerging applications, fostering a shared agenda for the age of multi-agent generative AI systems.

Call for Papers

We warmly invite submissions from researchers, practitioners, and students working at the intersection of multi-agent learning and generative AI. Accepted papers will be presented as posters, with a selection of outstanding submissions invited for spotlight or oral presentations. The workshop is non-archival.

Topics of Interest (including but not limited to)

• Multi-Agent Learning Paradigms for LLMs
  • Cooperative multi-agent reinforcement learning for improving coordination between modules within the multi-agent system LLMs (LLM orchestration)
  • Adversarial training for improving the generalizability of the single LLM training
  • Open multi-agent reinforcement learning/ad hoc teamwork for a multi-agent system LLMs to deal with some unknown and situational function/data providers
  • Formalism of full/partial information required in modelling multi-agent system LLMs and the minimal information each agent needs
  • Strengths/weaknesses of natural language as both the action space and the observation space in the multi-agent system LLMs
  • Criteria for evaluating a "well structured" multi-agent system LLMs in completing a task (e.g., game-theoretic approaches and models)
  • Coordination mechanisms for improving performance of multi-agent system LLMs, which can be either predefined or learned from data
  • Structures (e.g., chains, graphs, etc.) to represent a multi-agent system for LLMs
  • Application of coordination graphs (e.g., DAGs, factor graphs, etc.) on decomposing reward functions for training multi-agent system LLMs
• Generative AI for Multi-Agent Learning
  • World models for improving the data quality for multi-agent learning
  • Reward models for improving multi-agent learning with sparse rewards
  • Generative AI to generate a diverse set of agent models
  • Generative models (e.g., diffusion models) for improving multi-agent learning
  • Graph-based generative AI for improving graph-structured multi-agent learning and emergent communication between agents
  • Multi-agent systems for the modular generative models
• Multi-Agent Exploration for Generative AI
  • Multi-agent exploration for coordinating modules in the modular generative AI models
  • The role of entropy of agent policies (generators) in the modular generative AI learning
• Environments for Testing and Developing Multi-Agent Learning
  • Environments of real-world decentralised or distributed control problems
  • Computationally efficient environments for generative AI-based multi-agent learning
  • Light environments (without LLMs) for simulating the human-AI interaction process
  • JAX environments for accelerating multi-agent simulation processes
• Human-AI Interaction
  • Learning paradigms for improving the capability of AI agents to adapt to human instructions or proactively guide humans
  • Capable and interpretable (explainable) human models trained by generative AI
  • Appropriate medium of conveying human instructions to AI agents (e.g., natural language, formal methods and learning embeddings)
  • Approaches for estimating human intentions enabling AI agents to make better decision

Important Dates and Information

• Paper Submission Deadline: 5 February 2026, 11:59pm AOE 10 February 2026, 11:59pm AOE
• Notification of Acceptance (hard deadline): 1 March 2026, 11:59pm AOE
• Camera-Ready Due: 3 April 2026, 11:59pm AOE

Submission Platform: All submissions will be managed through OpenReview.

Note: Submissions are required to use the provided workshop LaTeX template (download here). Double-blind review policy applies.

Review Process

• Submission Platform: All submissions will be managed through OpenReview.
• Double-Blind Review: We enforce a double-blind review policy to ensure anonymity and impartiality for both authors and reviewers.
• Reviewer Assignment: Each submission will be evaluated by at least two members of the program committee with expertise in multi-agent learning, generative AI, or related fields.
• Final Decisions: Final acceptance decisions will be made by the organizing committee based on reviewer feedback and a thorough discussion period.
• Conflict of Interest: We will strictly adhere to the ICLR policy on Conflicts of Interest (COI). Reviewers will be required to declare any potential conflicts, and conflicted papers will be reassigned.
• LLM Usage Policy: We will follow the official ICLR 2026 Policies on Large Language Model Usage. In particular, AI systems may be used by workshop participants or organizers for supportive roles (e.g., language refinement, brainstorming, or discussion moderation), but AI-generated papers are not permitted for normal or tiny paper submissions. All AI contributions must remain under human oversight and validation, and the role of AI (if any) in the preparation of submissions must be transparently acknowledged.

Keynote Speakers

Yali Du

King's College London

Eugene Vinitsky

New York University

Zhijing Jin

University of Toronto

Natasha Jaques

University of Washington, Google DeepMind

Peter Stone

University of Texas at Austin, Sony AI

Panel Discussion

Topic: Bridging the Gap between Multi-Agent Learning and Generative Agents

Elias Stengel-Eskin

University of Texas at Austin

Eugene Vinitsky

New York University

Marc Lanctot

Google DeepMind

Bo An

Nanyang Technological University

Vincent Sunn Chen

Snorkel AI

Research Fellow / Founding Team

Schedule

Organizers

Jianhong Wang

University of Bristol

Caroline Wang

University of Texas at Austin, Google DeepMind

Feng Chen

Nanyang Technological University

Muhammad Arrasy Rahman

University of Texas at Austin

Felipe Leno da Silva

Lawrence Livermore National Laboratory

Rupali Bhati

Northeastern University

Bo Liu

National University of Singapore

Mustafa Mert Çelikok

University of Southern Denmark

Contact

For any inquiries, please contact iclr2026malgai@gmail.com. To catch up with us, please follow us on X.

Sponsor