Title: Optimization framework and sensitivity analysis of Stackelberg MFGs
Abstract: Stackelberg game is a bi-level game in which the leader chooses an optimal action in anticipation of the corresponding Nash equilibria (NE) from the followers, who would play the game according to any action taken by the leader. It is a general paradigm that can be used to model various economics and engineering problems, including ridesharing and principal-agent problems. Unlike the principal-agent problem, the general Stackelberg game may be unstable when the interests of multiple parties are not aligned and when there are multiple Nash equilibria choices for followers. In this talk, we will adopt an optimization framework for the characterization of Nash equilibria set and show that Stackelberg games are robust with proper relaxation. Based on joint work with Anran Hu (University of Oxford) and Jiacheng Zhang (UC Berkeley).