Title: A Reinforcement Learning Approach for Non-Truthful Mechanism Design
Abstract: Although the mechanism design literature has mainly focused on truthful mechanisms, most mechanisms used in practice are not truthful. Strategy-aware mechanism design is becoming critical, especially given the increasing evidence that algorithms operating in market platforms may adopt complex strategic behaviors. In this talk, I will introduce a new framework for non-truthful mechanism design where mechanisms are leader actions in a Stackelberg game between the mechanism designer and the participants. I will then present a reinforcement learning approach for mechanism design that builds on this framework. I will conclude by showing how this approach can be used to learn sequential mechanisms with limited communication and e-commerce platforms that prevent collusive behaviors of AI pricing algorithms.
