ƽÌØÎå²»ÖÐ

Ìý

Abstract
Game theoretic models are a popular framework for analyzing strategic interaction in systems with several autonomous agents. In this talk, we consider stochastic games and N-player mean-field games, two models that are useful for studying multi-agent learning. In this context, we discuss common learning paradigms used for selecting policies and give special attention to simple algorithms that can be used under decentralized information. We will present structural results on policy dynamics and illustrate how such results can guide algorithm design. Finally, we present a decentralized learning algorithm and analyze its convergence to near equilibrium policies in several classes of games.Ìý

Ìý

Bio: Bora Yongacoglu is a post-doctoral fellow in the department of mathematics and statistics at Queen's University. He recently received his PhD in mathematics at Queen's University, where he was supervised by Serdar Yuksel and Gurdal Arslan. Before Queen's, he studied economics and mathematics at ƽÌØÎå²»ÖÐ. His research focuses on multi-agent reinforcement learning and strategic dynamics in games.Ìý