Prof. Jeff Shamma, University of Illinois Urbana-Champaign

Title: Multi-Agent Learning vs Equilibrium in Games

Abstract: The framework of multi-agent in games learning explores the dynamics of how individual agent strategies evolve in response to the evolving strategies of other agents. Of particular interest is whether agent strategies converge to well-known solution concepts, such as Nash Equilibrium (NE) or its variants, under natural assumptions on the learning strategies. This talk begins with a brief introduction of the learning in games framework and an overview of both positive and negative results that describe when specific learning algorithms do or do not converge to NE, respectively. The talk concludes with a presentation of recent results that illustrate how a system theoretic approach to learning in games can leverage feedback control concepts (e.g., decentralized stabilization, strong stabilization, and robustness) to lead to new perspectives on what is or is not achievable in learning in games.