Prof. Haifeng Xu - Algorithmic Information Design: Computability, Robustness, and Learnability

The celebrated field of mechanism design studies how a system designer can design agents' incentives, and consequently their actions, in order to steer their joint decisions towards a desirable outcome. This talk also examines the intervention of agents' actions but through a fundamentally different yet equally important "knob" --- i.e., influencing agents' decisions by designing the available information to each agent. This task, a.k.a. information design, is particularly relevant in today's digital economy and has found numerous applications such as auction design, ride sharing systems, preference or information elicitation. Information design has attracted explosive recent interest in economics and computer science. This talk will examine a foundational model in this space, namely, the Bayesian persuasion (BP) model. Like mechanism design, it is intrinsically an algorithm design problem subject to incentive constraints. We will present a relatively complete set of algorithmic results about BP, including its computability, robustness and learnability. /***********************************************/ Haifeng Xu is the Alan Batson Assistant Professor in Computer Science at the University of Virginia and a visiting research scientist at Google. He studies decision making and machine learning in multi-agent environments, particularly in informationally complex setups (e.g., with asymmetric or limited access to information/data). Prior to UVA, Haifeng was a postdoc at Harvard and obtained his PhD in Computer Science from the University of Southern California. His research has been recognized by multiple awards, including a Google Faculty Research Award, honorable mention for the ACM SIGecom Dissertation Award, runner-up for the IFAAMAS Victor Lesser Distinguished Dissertation Award, a Google PhD fellowship, and multiple best paper awards.