机器学习与数据科学博士生系列论坛（第八十五期）—— Dynamic Contract Design in Principal-Agent Markov Decision Process
报告人： 陈雨静（北京大学）
时间：2025-03-20 16:00-17:00
地点：腾讯会议 531-8098-3912
Abstract:
The principal-agent problem, a canonical model in algorithmic game theory, has witnessed heightened theoretical significance due to the emergence of self-interested learning agents in decentralized decision-making environments. Contract theory operationalizes incentive alignment through outcome-contingent payment schemes, but struggles to adapt dynamically when agents continuously learn and optimize their strategies. Reinforcement learning (RL) offers a powerful solution by enabling the automated design of adaptive contracts that optimize incentives over time.
We introduce recent work that integrate RL into principal-agent settings by formulating the problem as a Principal-Agent Markov Decision Process (MDP) with dynamic contract adaptation. We further propose a Markovian policy framework to achieve subgame perfect equilibrium and introduce non-Markovian policy to address myopic incentive constraints, expanding the scope of dynamic contract design in complex decision environments.
About the Speaker:
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。
Your participation is warmly welcomed!
欢迎扫码关注北大统计科学中心公众号，了解更多讲座信息!
