The Berkeley–Hopkins Seminar Series on Learning-Enabled Multi-Agent Systems (LEMAS) is a weekly seminar that brings together researchers working at the forefront of learning and/or incentive design in multi-agent systems. We hope to create a platform for sharing recent advances, discussing key challenges, and fostering new research directions. We are based at Johns Hopkins University and UC Berkeley and currently operating in a hybrid format (talks occur both in-person and livestreamed over Zoom).
Embedded Files
Time: Every Friday, 10-11 a.m. Pacific Time
Zoom link for the virtual attendance: https://wse.zoom.us/j/97690026042
Organizers:
Chinmay Maheshwari (Assistant Professor, ECE, Johns Hopkins University)
Shankar Sastry (Professor, EECS, UC Berkeley)
Pan-Yang Su (Ph.D. Student, EECS, UC Berkeley)
Youtube Channel: https://www.youtube.com/@Berkeley-HopkinsLEMAS
Spring 2026 Schedule
Spring 2026 Schedule
Note: Presentation titles in parentheses provide tentative descriptions of upcoming talks, and will be replaced with actual presentation titles when available. Titles for past talks link to more information about the talk and speaker, as well as a recording and/or slide deck when available.
Date Speaker Presentation Title
February 13 Haifeng Xu The Interplay of Economic Thinking and Language Models: Vignettes and Lessons
_______________________________________________________________________________________________________________________________
February 27 Gabriele Farina How Strong a Notion of Rationality Can We Learn Efficiently in Multi-Agent Settings?
_______________________________________________________________________________________________________________________________
March 6 Mingyan Liu Inter-temporal Incentive and Its Application in Strategic Classification and Security Audit
_______________________________________________________________________________________________________________________________
April 3 Milind Tambe Generative AI for Global Social Impact: Towards Solving the Deployment Bottleneck
_______________________________________________________________________________________________________________________________
April 10 Amanda Prorok Trustworthy Robot Teams
_______________________________________________________________________________________________________________________________
April 17 Ketan Savla Towards Frugal Mechanism Design for Routing Games
_______________________________________________________________________________________________________________________________
April 24 Stephanie Gil Resilient Multi-Robot Systems: Coordination and Learning in the Real-World
_______________________________________________________________________________________________________________________________
May 8 Xiaowu Dai Training-Free Multi-Agent Language Models via Mechanism Design
_______________________________________________________________________________________________________________________________
May 15 Alireza Fallah
TalK information (Spring 2026)
TalK information (Spring 2026)
Haifeng Xu (UChicago)
Title: The Interplay of Economic Thinking and Language Models: Vignettes and Lessons
Abstract: Although large language models (LLMs) have achieved remarkable technological successes, their effective application in economic domains—critical for realizing their broader promise across society and markets—remains largely underexplored. This talk presents our recent research efforts at bridging economic thinking and LLMs, with concrete instantiations in areas such as mechanism design, information design, and forecasting. A key theme of the talk is to illustrate how this emerging technology gives rise to genuinely new research questions in foundational economic settings. We hope that explorations along this line will help move LLM research towards realizing its true economic impact in the real world.
Bio: Haifeng Xu is an assistant professor in computer science and data science at the University of Chicago, and directs the Strategic IntelliGence for Machine Agents (SIGMA) research lab which focuses on multi-agent AI and data-driven decision making. His works have been recognized by a few major awards including AI2050 Early Career Fellow, IJCAI Early Career Spotlight, a Best Paper at the Web Conference and a Best Student Paper at AAMAS. Haifeng publishes regularly at leading venues in machine learning and computational economics, and serves as area chair or senior program committee for venues such as ICML, NeurIPS, EC, AAAI, etc.
Video Link: https://youtu.be/TnYnkL1ERdo
Gabriele Farina (MIT)
Title: How Strong a Notion of Rationality Can We Learn Efficiently in Multi-Agent Settings?
Abstract: A central question in multi-agent learning is: how strong a notion of rational behavior can be guaranteed through efficient no-regret learning? Classical results show that minimizing regret leads to coarse correlated equilibria, and enriching the class of allowed deviations yields progressively stronger equilibrium concepts. But how far can this hierarchy be pushed while retaining efficient algorithms in general convex and extensive-form games?
In this talk, I present recent results that characterize the strongest notions of rationality that can be efficiently learned in this broad setting. We show that linear and low-degree polynomial correlated equilibria—strictly stronger than coarse correlated equilibria and natural relaxations of correlated equilibria—can be computed and learned efficiently even in general convex and extensive-form games.
En route to the result, we will introduce new algorithmic tools with broader applicability. First, the natural deviation sets underlying stronger notions of regret do not admit efficient separation or optimization oracles. To address this, we introduce a new algorithmic primitive, semiseparation, which enables regret minimization over convex sets that lack classical separation oracles. Second, our analysis leverages a fast computational version of von Neumann’s minimax theorem, yielding convergence rates that scale logarithmically in the desired accuracy. This tool has further applications, including to variational inequalities, fixed-point computation, and online multicalibration.
Based on joint work that has appeared at STOC’25 and ACM EC’25.
Bio: Gabriele Farina is an Assistant Professor in the Department of Electrical Engineering and Computer Science at MIT, with affiliations in LIDS and the Operations Research Center. His research combines techniques and notions of strategic behavior from game theory together with modern tools from machine learning, optimization, and statistics to construct state-of-the-art methods for learning and decision-making in multi-agent systems. He received his Ph.D. in Computer Science from Carnegie Mellon University, and his work has been recognized with several awards, including a Best Paper Award at NeurIPS'20 and an Outstanding Paper Honorable Mention at ICLR'23. His dissertation received the ACM SIGecom Doctoral Dissertation Award and one of the two ACM Dissertation Award Honorable Mentions, among others. He is an AI2050 Early Career Fellow, and the recipient of an NSF CAREER award.
Video link: https://www.youtube.com/watch?v=X5vd5iS6_WY
Mingyan Liu (U Mich)
Title: Inter-temporal Incentive and Its Application in Strategic Classification and Security Audit
Abstract: Strategic classification problems are typically framed as a Stackelberg game, where self-interested individuals or agents manipulate their response to obtain favorable decision outcomes made by a classifier, generally turning to dishonest actions when they are less costly than genuine efforts. It is now well understood that a one-shot formulation is fundamentally unable to mitigate this, whereas repeated interactions give rise to what's known as inter-temporal incentives that can give the designer (the principal) greater ability to induce honesty. In this talk I will present two sequential models that explore the strategic principal-agent interactions in such a context. In the first case, an agent has to go through a progression of classifiers with increasing rewards but also increasing difficulty. I will characterize the agent's best response to a given progression, and derive conditions under which the principal, through judicious design of the sequence of classifiers, can incentivize an agent to reach an arbitrarily high level via solely honest means. In the second case, the agent does not have a cheating option but can choose to opt out of the classification exercise entirely. I will examine how the designer might incentivize the agent to opt in and to exert higher effort. (This is joint work with my former student Ziyuan Huang and current student Lina Alkami)
Bio: Mingyan Liu is the T.C. Chang Professor of Engineering, a professor of Electrical Engineering & Computer Science, and the Associate Dean for Academic Affairs of the College of Engineering at the University of Michigan, Ann Arbor. She received her Ph.D. Degree in electrical engineering from the University of Maryland, College Park, in 2000 and has been with the University of Michigan ever since. Her research interests are in resource allocation, sequential decision and learning theory, game theory and incentive mechanisms, with applications to large-scale networked systems, cybersecurity and cyber risk quantification. Some of her research in this space has been successfully commercialized. She is a Fellow of the IEEE and a member of the ACM.
Video Link: https://www.youtube.com/watch?v=wMGwPBQGS3k
Milind Tambe (Harvard)
Title: Generative AI for Global Social Impact: Towards Solving the Deployment Bottleneck
Abstract: My team’s work on AI for Social Impact (AI4SI) has spanned two decades, focusing on optimizing limited resources in critical areas like public health, conservation, and public safety. I will present field results from India, where the deployment of restless and collaborative multi-armed bandit (RMAB) algorithms achieved significant improvements in the world’s two largest mobile maternal health programs. I will also highlight ongoing work on network-based HIV prevention in South Africa, modeled as a branching bandit problem. These projects, along with other initiatives across Africa and Asia, expose a critical "deployment bottleneck" that spans the entire machine learning pipeline. This bottleneck consists of three key hurdles: the observational scarcity gap (data), the policy synthesis gap (learning and modeling), and the human-AI alignment gap (deployment). This talk investigates how Generative AI can address this AI4SI deployment bottleneck through the strategic use of LLM Agents and diffus models. I will demonstrate how LLM Agents address the alignment gap by integrating expert guidance into algorithmic planning, ensuring resource optimization strategies reflect real-world priorities. Furthermore, I will show how diffusion models mitigate the scarcity and synthesis gaps by generating synthetic social networks and facilitating Transfer RL to utilize data across domains. I will conclude by discussing this path toward a more scalable, human-aligned future for AI for Social Impact.
Bio: Milind Tambe is the Gordon McKay Professor of Computer Science at Harvard University; concurrently, he is Principal Scientist and Director for “AI for Social Good” at Google Research. Prof. Tambe and his team have developed innovative AI and multi-agent reasoning systems that have been successfully deployed to deliver real-world impact in public health (e.g., maternal and child health), public safety, and wildlife conservation. He is the recipient of the AAAI Award for Artificial Intelligence for the Benefit of Humanity, the AAAI Feigenbaum Prize, the IJCAI John McCarthy Award, the AAAI Robert S. Engelmore Memorial Lecture Award, and the AAMAS ACM/SIGAI Autonomous Agents Research Award. He is a fellow of AAAI and ACM. His contributions in Operations Research and public safety have also been recognized with the INFORMS Wagner Prize for excellence in Operations Research practice, Military Operations Research Society Rist Prize, the Columbus Fellowship Foundation Homeland security award, and commendations and certificates of appreciation from the US Coast Guard, the Federal Air Marshals Service, and airport police at the city of Los Angeles.
Amanda Prorok (University of Cambridge)
Title: Trustworthy Robot Teams
Abstract: As autonomous agents transition into our physical reality, the mandate for trustworthy AI must expand to address new real-world challenges. This talk explores our research across several key pillars of this transition, moving from formal safety guarantees to the long-term accountability of collective robot teams. The narrative begins with the foundation of safety and robustness: I will discuss how we provide provable safety guarantees in dynamic environments and how we co-design physical spaces to mitigate the risks of real-world execution uncertainty. We then scale these principles to collective intelligence, demonstrating how behavioral diversity acts as a catalyst for "latent resilience," allowing large-scale robot teams to maintain performance even under significant disruption. Finally, the talk culminates in the pillar of accountability. I will introduce a novel method for remote policy watermarking that embeds detectable signatures into a robot’s motion, effectively bridging the gap between physical action and digital provenance. Together, these pillars establish a technical and ethical framework to deploy responsible, scalable robot teams in human-shared environments.
Bio: Amanda Prorok is Professor of Collective Intelligence and Robotics in the Department of Computer Science and Technology at the University of Cambridge, and a Fellow of Pembroke College. In her work, she pioneered differentiable communications methods for multi-agent systems, with applications to multi-robot perception and control. Amanda has given invited keynotes at TEDx and IEEE ICRA and has received numerous research awards, including a prestigious ERC Starting Grant. Amanda is an IEEE Senior Member, serves as Senior Editor for IEEE Transactions on Robotics (T-RO) and chaired the 2021 IEEE International Symposium on Multi-Robot and Multi-Agent Systems. Her PhD thesis received the Asea Brown Boveri (ABB) prize for the best thesis in Computer Science at EPFL.
Ketan Savla (University of Southern California)
Title: Towards Frugal Mechanism Design for Routing Games
Abstract: Congestion pricing has been a key tool to distribute traffic across time and space, but effective pricing depends on individual attributes—such as value of time—which vary by traveler, traffic conditions, and economic context. Traditional stated-preference methods are too coarse and resource-intensive to keep these estimates accurate. Meanwhile, non-monetary approaches like travel recommendations remain uncoordinated and can lead to suboptimal (inverse U-shaped) social welfare outcomes.
We propose coordinated private routing recommendations and adaptive pricing to address these challenges. We show that information platforms able to link individual decisions to societal regret can obviate the need for pricing to steer collective behavior to desired outcome. Through controlled human-subject experiments and case studies, we demonstrate that the proposed mechanisms can achieve social objectives while requiring fewer resources from both agencies and travelers than conventional approaches.
Bio: Ketan Savla is an Associate Professor and the John and Dorothy Shea Early Career Chair in Civil Engineering at the University of Southern California. His current research interest is in distributed optimal and robust control, dynamical networks, state-dependent queuing systems, and mechanism design, with applications in civil infrastructure and autonomous systems. His recognitions include NSF CAREER, an IEEE CSS George S. Axelby Outstanding Paper Award, the AACC Donald P. Eckman Award, and the IEEE ITSS Outstanding Application Award. He is also a co-founder and the chief science officer of Xtelligent, Inc.
Stephanie Gil (Harvard University)
Title: Resilient Multi-Robot Systems: Coordination and Learning in the Real-World
Abstract: Multi-robot systems are increasingly integrated into real-world applications, from autonomous vehicle fleets to search-and-rescue teams. Ensuring their coordination algorithms remain robust against unreliable communication, security threats, and corrupted data is essential. We explore how control and information exchange can enhance situational awareness and security in multi-robot networks. For example, consensus problems are at the core of many multi-agent coordination tasks -- a key challenge is that classical results fail when malicious agents exceed half of the network connectivity. This quickly leads to limitations in the practicality of many multi-robot coordination tasks. However, with the growing prevalence of cyber-physical systems comes novel opportunities for detecting attacks by using cross-validation with physical channels of information. We introduce the concept of stochastic observations of trust, where an agent’s trustworthiness is modeled probabilistically. Under this framework, consensus can be restored even when the number of malicious agents surpasses classical limits. We will present both theoretical insights and experimental results to this end. Additionally, we will briefly discuss reinforcement learning approaches that fuse real-time sensor data with learned policies for sequential decision-making in dynamic, partially observable environments. Case studies include autonomous delivery routing and real-time rendezvous with sperm whales in Dominica. Together, these advances move toward multi-robot systems that are both secure and capable for the real-world.
Bio: Stephanie is the John L. Loeb Associate Professor of Engineering and Applied Science (SEAS) and an Associate Faculty member of the Kempner Institute at Harvard University. Her research focuses on trust and coordination in multi-robot systems, with applications in security, communication, and autonomy. Her contributions to the field have been recognized through the DARPA Young Faculty Award (2024), the Office of Naval Research Young Investigator Award (2021), and the National Science Foundation CAREER Award (2019). She was also named a 2020 Sloan Research Fellow for her work at the intersection of robotics and communication. She earned her Ph.D. from CSAL at MIT, specializing in multi-robot coordination and control, and completed her B.S. at Cornell University.
Xiaowu Dai (UCLA)
Title: Training-Free Multi-Agent Language Models via Mechanism Design
Abstract: Large Language Models (LLMs) have demonstrated strong generative capabilities but remain prone to inconsistencies and hallucinations. We introduce Peer Elicitation Games (PEG), a training-free, game-theoretic framework for aligning LLMs through a peer elicitation mechanism involving a generator and multiple discriminators instantiated from distinct base models. Discriminators interact in a peer evaluation setting, where rewards are computed using a determinant-based mutual information score that provably incentivizes truthful reporting without requiring ground-truth labels. We establish theoretical guarantees showing that each agent, via online learning, achieves sublinear regret in the sense their cumulative performance approaches that of the best fixed truthful strategy in hindsight. Moreover, we prove last-iterate convergence to a truthful Nash equilibrium, ensuring that the actual policies used by agents converge to stable and truthful behavior over time. Empirical evaluations across multiple benchmarks demonstrate significant improvements in factual accuracy. These results position PEG as a practical approach for eliciting truthful behavior from LLMs without supervision or fine-tuning.
Bio: Xiaowu Dai is an assistant professor in the Departments of Statistics and Data Science, and of Biostatistics at UCLA. Before joining UCLA, he did a postdoc at UC Berkeley working with Prof. Michael Jordan, and received a Ph.D. in Statistics at UW-Madison advised by Prof. Grace Wahba. His research mainly focuses on statistical theory and methodology for real-world problems that blend computational, inferential, and economic considerations.
Page updated
Report abuse