Seminar: Stochastic Principal-Agent Problems: Efficient Computation and Learning by Dr. Jiarui Gan

Abstract:  The principal-agent problem is an economic framework for studying the challenges of aligning incentives between a principal and self-interested agents, especially under information asymmetry. This paradigm has a broad range of applications from public decision-making and transportation to e-commerce and financial services. There has been a recent surge of interest in exploring principal-agent interactions in stochastic and dynamic environments, where the players' decisions unfold over a sequence of time steps. In this talk, I will introduce a framework of stochastic principal-agent problems and discuss our recent algorithmic results. I will present an efficient algorithms for computing and learning the principal’s optimal coordination policies in this framework. The algorithms are based on a novel value-set-iteration method, which allows policies to be contingent on the interaction history. The policies obtained therefore offer strict utility improvement over stationary policies while being computationally more amenable. These results have direct implications to related fields including stochastic games and multi-agent learning.

Bio:  Jiarui Gan is a Departmental Lecturer at the Computer Science Department, University of Oxford, working in the Artificial Intelligence & Machine Learning research theme. Before this he was a postdoctoral researcher at Max Planck Institute for Software Systems, and he obtained his PhD from Oxford. Jiarui is broadly interested in algorithmic problems in game theory. His current focus is on sequential principal-agent interactions in stochastic environments. His recent work has been selected for an Outstanding Paper Honorable Mention at the AAAI'22 conference.