Utilizing deep learning and game theory to find optimal policies for a large number of noncooperative agents
Gökçe Dayanikli
Assistant Professor, Department of Statistics, University of Illinois Urbana-Champaign
Abstract: In this talk, we will discuss how we can utilize deep learning to solve complex (dynamic and stochastic) game theoretical problems where there many agents (such as banks, companies, or people) interacting. We will first look at a stochastic optimal control problem for one agent and explain how we can use deep learning to solve this problem. Later, we will move on to the multi-agent setup, and we will discuss and compare two equilibrium notions in game theory: Nash equilibrium and Stackelberg equilibrium. After explaining how a Nash equilibrium can be approximated for dynamic and stochastic games with a large number of agents through mean field games, we will introduce the Stackelberg mean field games between a principal (i.e., regulator) and many agents. Stackelberg mean field game models can be used to find incentives or optimal policies for a large group of noncooperative agents with a motivation to model different real-life problems such as regulating the systemic risk in banking sector or mitigating an epidemic. We will discuss how (intrinsically bi-level) Stackelberg mean field game model can be rewritten to propose a single-level deep learning method to solve this complex problem and conclude with some examples.
Bio: Gökçe Dayanikli is an assistant professor at the University of Illinois Urbana-Champaign, Department of Statistics. Before joining UIUC, she was a term assistant professor of Statistics at Columbia University. She completed her Ph.D. in Operations Research & Financial Engineering at Princeton University where she was awarded the School of Engineering and Applied Sciences Award for Excellence. During Fall 2021, she was a visiting graduate researcher at the Institute for Mathematical and Statistical Innovation (IMSI) to participate in the "Distributed Solutions to Complex Societal Problems" program. Her research interests are analysis and applications of Mean Field Games & Control and Graphon Games and numerical approaches to these problems.