Multi-Agent Learning in Games

    13:30 - 14:30, Dec. 21, 2024, UTC+8
    Georgios Piliouras
    About the Speaker
    Georgios Piliouras is game theory team lead at Google DeepMind and associate professor at the Singapore University of Technology and Design (SUTD). His research interests lie in the areas of multi-agent learning, algorithmic game theory, blockchain, and dynamical systems. He received his PhD in Computer Science from Cornell University. He has held academic positions at Georgia Institute of Technology, California Institute of Technology, and UC Berkeley. He is a research affiliate of the Ethereum Foundation. He is the recipient of a Singapore NRF Fellowship and a Simons Fellowship. His work has been recognized with multiple top honors and awards at premier conferences on Machine Learning, AI, multi-agent systems, and blockchain, including in venues such as ICLR, AAAI, AAMAS, and IEEE Blockchain.
    Abstract
    We examine some classic questions in game theory and online learning. How do standard learning dynamics such as multiplicative weights update, gradient descent, and variants thereof behave when applied in games? The traditional approach to this question is to connect it to different notions of game theoretic equilibria. We discuss why such results can be uninformative in practice and instead offer new insights by using tools and techniques from dynamical systems such as chaos theory, recurrence, and conservative systems.