In recognition of Professor Tamer Başar’s pioneering contributions to control theory, game theory, and decision-making and in celebration of his 80th birthday, a one-day workshop will be held on Friday, September 19, 2025, as part of the 61st Annual Allerton Conference. Attendance is free for all Allerton Conference participants.
A detailed program for the workshop can be found below.
Friday, September 19, 2025
| TIME | EVENT |
| 9:00 – 9:15 a.m. | Welcome and Opening Remarks National Center for Supercomputing Applications (Auditorium: 1122NCSA) |
| 9:15 – 10:45 a.m. | Session 1 Session Chair: Jeff Shamma (UIUC) National Center for Supercomputing Applications Building (1122 NCSA) 9:15-9:45 a.m. Peter Caines (McGill University)  Title: Control and Games on Large Sparse Networks: from Graphons to Graphexons 9:45-10:15 a.m. Asu Ozdaglar (MIT)  Title: Optimizing Data for Decision-Making Abstract: We study the fundamental question of how informative a dataset is for solving a given decision-making task. In our setting, the dataset provides partial information about unknown parameters that influence task outcomes. Focusing on linear programs, we characterize when a dataset is sufficient to recover an optimal decision, given an uncertainty set on the cost vector. Our main contribution is a sharp geometric characterization that identifies the directions of the cost vector that matter for optimality, relative to the task constraints and uncertainty set. We further develop a practical algorithm that, for a given task, constructs a minimal or least-costly sufficient dataset. Our results reveal that small, well-chosen datasets can often fully determine optimal decisions—offering a principled foundation for task-aware data selection. 10:15-10:45 a.m. Shankar Sastry (UC Berkeley)  Title: Learning Enabled Multi-Agent Systems in Societal Systems Transformation Abstract: The heart of the digital transformation of societal systems is the development of a Theory of Learning Multi-Agent Systems. The transformation of societal systems in energy, transportation, health care, manufacturing and financial systems is accompanied by issues of economic models, privacy, security and fairness considerations. Indeed, the area of “mechanism design” for societal scale systems is a key feature in transitioning the newest technologies and providing new services. Crucially, human beings interact with automation and change their actions in response to incentives offered to them. This is where data driven methods of Machine Learning can be applied to multi-agent systems can help shape how they dynamically adapt to changing circumstances. In this talk, I will present a few vignettes: how to align societal objectives with Nash equilibria using suitable incentive design, proofs of stability of decentralized decision making while learning preferences and a new take on Multi-Agent RL algorithms. The application of the techniques to problems in road transportation, air transportation (such as drones for air taxis and humanitarian assistance will be presented. Recently, we have recently been applying these methods to fields of robotics and embodied intelligence such as in robotic Indy car racing and table tennis. These applications will also be presented. The work is joint with Chinmay Maheshwari, Manxi Wu, Kshitij Kulkarni, Pan Yang Su, and Dvij Kalaria. |
| 10:45 – 11:00 a.m. | Coffee Break |
| 11:00 a.m. – 12:30 p.m. | Session 2 Session Chair: Bahman Gharesifard (Queen’s University) National Center for Supercomputing Applications Building (1122 NCSA) 11:00-11:30 a.m. Miroslav Krstic (University of California San Diego)  Title: Nash Feedback Strategies Without Isaacs PDEs for a Driftless Vehicle Model Abstract: For Tamer’s special milestone I try to offer a result perhaps commensurate with the magnitude of his influence over my research. Global stabilizers for nonholonomic cars in non-Cartesian coordinates, developed this summer, open the door to inverse optimal games. For parking, these controllers minimize infinite horizon penalties on the position and heading states (non-quadratic) and speed and steering inputs (quadratic). When a disturbance acts on steering (wheel misalignment, uneven tire pressure, worn steering rack bushings, etc.), we solve a zero-sum game with penalized disturbance. The feedbacks designed – gradients of our Lyapunov functions which solve the HJB/HJI PDEs by construction – are Nash strategies no matter how low or high their (positive) gains. Such robustness is generically unachievable even for disturbance-free linear systems (LQR’s gain margin is 1/2 to infinity). Gain-invariant results hold here because (1) the nonholonomic vehicle is a DRIFTLESS system and (2) our Lyapunov functions for it are STRICT. The problem of strict CLF construction for nonholonomic systems was open for 40+ years. It wasn’t only stability analysis which the need to employ non-quantitative LaSalle arguments had hampered for decades. The achievement of optimality, gain-robustness, and disturbance-robustness was also impossible. They are all now achieved, courtesy of strict CLFs. 11:30 a.m.-12:00 p.m. Murat Arcak (UC Berkeley)  Title: Data-Driven Reachable Set Estimation Abstract: The computation of reachable sets is essential for characterizing and verifying the behavior of safety-critical systems. However, many practical systems are high-dimensional and analytically intractable, making the exact computation of reachable sets difficult or impossible. We propose a data-driven approach that uses a finite ensemble of sample trajectories to estimate reachable sets with probabilistic accuracy guarantees. This approach is broadly applicable and computationally advantageous, as the main cost comes from simulating a predetermined number of trajectories, which can be parallelized to reduce computation time. 12:00 p.m.-12:30 p.m. Geir Dullerud (University of Minnesota)  Title: Deploying Large Language Models (LLMs) in Control Design: ControlBench and ControlAgent Abstract: This presentation explores the convergence of large language models (LLMs) and machine reasoning, with an emphasis on their applications in control engineering. Contemporary LLMs—advanced foundation models equipped with extensive knowledge bases—have demonstrated significant potential in tackling complex reasoning and programming challenges. But what value can they offer control engineers? We will discuss our recent work on exploring the capabilities of state-of-the-art large language models (LLMs), such as GPT-4, o4, Claude 3 Opus, and Gemini 1.0 Ultra, in solving control design problems. In particular, we introduce ControlBench, a benchmark dataset tailored to reflect the breadth, depth, and complexity of classical control design. We also present ControlAgent, a new paradigm that automates control system design via novel integration of LLM agents and control-oriented domain expertise. ControlAgent encodes expert control knowledge and emulates human iterative design processes by gradually tuning controller parameters to meet user-specified requirements for stability, performance, and robustness. This study serves as an initial step towards the broader goal of employing artificial general intelligence in control engineering. |
| 12:30 – 2:30 p.m. | Lunch Break |
| 2:30 – 4:30 p.m. | Session 3 Session Chair: Venugopal Veeravalli (UIUC) National Center for Supercomputing Applications Building (1122 NCSA) 2:30-3:00 p.m. John Bailleul (Boston University)  Title: The Game is Afoot: Thoughts on the Intellectual Legacy of Tamer Basar 3:00-3:30 p.m. Vincent Poor (Princeton University)  Title: Resource Constrained Learning over Wireless Networks Abstract: It is anticipated that the next generation of wireless networks will incorporate AI to a significant degree at all network layers. A major part of this trend is the migration of AI and machine learning functions to the network edge. There are several reasons for this: (i) a growing number of AI applications demand implementations involving end-user devices, (ii) much data of interest is collected at the network edge, and (iii) fog/edge computing has emerged to take advantage of the increasing sophistication of end-user devices. A notable framework for engaging the wireless network edge in machine learning is wireless federated learning, in which multiple end-user devices collaborate with the help of an aggregator to build a common model, each using its local data. In this framework, exchanges between end-user devices and the aggregator necessarily take place over wireless links. Since wireless networks are notoriously resource-limited, this creates a situation in which the interactions between the wireless medium and machine learning algorithms must be considered as a factor in the design and implementation of AI applications. This talk will explore aspects of this problem, including tradeoffs among energy consumption and other criteria such as bandwidth efficiency, learning rate and data privacy. 3:30-4:00 p.m. John Baras (University of Maryland)  Title: Risk-Sensitive Safety Filters for Reinforcement Learning with Probabilistic Guarantees Abstract: Humans have the ability to deviate from their natural behavior when necessary, which is a cognitive process called response inhibition. Similar approaches have independently received increasing attention in recent years for ensuring the safety of control. Realized using control barrier functions or predictive safety filters, these approaches can effectively ensure the satisfaction of state constraints through an online adaptation of nominal control laws, e.g., obtained through reinforcement learning. While the focus of these realizations of inhibitory control has been on risk-neutral formulations, human studies have shown a tight link between response inhibition and risk attitude. Inspired by this insight, we propose a flexible, risk-sensitive method for inhibitory control. Our method is based on a risk-aware condition for value functions, which guarantees the satisfaction of state constraints. We propose a method for learning these value functions using common techniques from reinforcement learning and derive sufficient conditions for its success. By enforcing the derived safety conditions online using the learned value function, risk-sensitive inhibitory control is effectively achieved. The effectiveness of the developed control scheme is demonstrated in simulations. 4:00-4:30 p.m. Francesco Bullo (University of California, Santa Barbara),  Title: Perspectives on Biologically Plausible Optimization Abstract: Biological neural circuits are often described as optimizing energy or cost functions, yet the precise nature of such optimization remains often elusive. In this talk, I present a systems-theoretic perspective on biologically plausible optimization, with a focus on sparse signal reconstruction as a case study. I will show how proximal gradient methods provide a principled way to transcribe convex optimization problems into recurrent neural network dynamics, yielding competitive firing-rate models with clear biological interpretations such as lateral inhibition. Using contractivity and non-expansiveness analysis, I establish sharp convergence guarantees for these dynamics, including linear–exponential convergence under restricted isometry conditions. This framework offers a normative, top-down explanation of neural circuit functionality, bridging neuroscience, control theory, and machine learning. |
