Abstract

While classic game theory often assumes partial, full, or common knowledge of game parameters, they are usually players' private information in the real world. Although machine learning can be a tool to learn them from data, rational agents may utilize this information advantage to gain profit. For a more robust application of machine learning and game theory into practice, it is crucial to study how to learn private information and how agents will utilize their information advantage in diverse and complex environments.

This talk will highlight our work on understanding, learning, and computing the optimal ways for agents to utilize their information advantage in games over artificial intelligence. I will mainly discuss two settings: (1) players' private payoff misreporting in Stackelberg equilibria and (2) coordinated auto-bidding algorithm design in online ad auctions.

I will conclude with several future directions on private information learning, protection, and counteraction.  

Time

2023-12-19  15:00 - 16:00   

Speaker

Yurong Chen, Peking University

Room

Room 308