![](../../images/mnav2.png)
机器学习与数据科学博士生系列论坛(第七十五期)—— Strategies in Repeated Games with Learning Agents
Speaker(s):陈雨静 (北京大学)
Time:2024-07-04 16:00-17:00
Venue:腾讯会议 627-5441-1672
摘要:
Autonomous learning agents are becoming increasingly integral to many online platforms, and economic agents can be modeled as learning agents. The concept of learning as a behavioral model traces back to the early economic literature on learning in games and has seen renewed interest within the computer science community. Instead of using a Nash equilibrium strategy as the best response, the agent can only base their actions on past information
As for such a playing game, what is the optimal strategy for an opponent playing against a learning agent? In a single-shot strategic game where learning is irrelevant, a Nash equilibrium becomes the reasonable prediction of the game’s outcome. However, the opponent can achieve higher utility in a repeated game over T rounds.
In this talk, we will first introduce the case in the normal-form game, focusing on no-regret learning agent and no-swap-regret learning agent. Then we will delve into the significance of swap regret for learning in game. Following this, we will discuss recent applications of these concepts in principal-agent problems, including auctions and Bayesian persuasion.
论坛简介:该线上论坛是由张志华教授机器学习实验室组织,每两周主办一次(除了公共假期)。论坛每次邀请一位博士生就某个前沿课题做较为系统深入的介绍,主题包括但不限于机器学习、高维统计学、运筹优化和理论计算机科学。