系综合学术报告第28期：
报告题目：An alpha-potential game framework for dynamic N-player games
报告人：Prof. Xin Guo (UC Berkeley)
时间：2026年6月10日（星期三）下午3:00-4:00
地点：清华大学双清综合楼304
摘要：Game theory has a long history and the min-max game has been well studied ever since Von Neumann and Nash. The leap from min-max (zero-sum) games to general-sum games is a fundamental escalation in computational and conceptual complexity. Over the past decade, mean field game theory has emerged as a pivotal framework, offering profound theoretical insights and computational advances for the analysis of large-scale, non-zero-sum stochastic games. However, mean field games require homogeneity and weak interaction among players and focus on the limiting behavior when N goes to infinity. In this talk we will present a new paradigm for dynamic N-player non-cooperative games called alpha-potential games, where the change of a player's value function upon unilateral deviation from her strategy is equal to the change of an alpha-potential function up to an error alpha. This game framework is shown to reduce the challenging task of finding alpha-Nash equilibria for a dynamic game to minimize the associated alpha-potential function. The latter is then shown to be a conditional McKean-Vlasov control problem. In such games, analysis of alpha reveals critical game characteristics, including choices of admissible strategies, the intensity of interactions, and the level of heterogeneity among players. We will discuss through simple examples some recent theoretical developments, their connections with mean-field games, along with a few open problems for this new game framework.
系综合学术报告第29期：
报告题目： Deterministic Policy Gradient for Reinforcement Learning with Continuous Time and Space
报告人：Prof. Xin Guo (UC Berkeley)
时间：2026年6月12日（星期五）下午3:00-4:00
地点：清华大学双清综合楼304
摘要: The theory of continuous-time reinforcement learning (RL) has progressed rapidly in recent years. While the ultimate objective of RL is typically to learn deterministic control policies, most existing continuous-time RL methods rely on stochastic policies. Such approaches often require sampling actions at very high frequencies, and involve computationally expensive expectations over continuous action spaces, resulting in high-variance gradient estimates and slow
convergence. In this talk, we will introduce deterministic policy gradient (DPG) methods for continuous-time RL. We will derive a continuous-time policy gradient formula expressed as the expected gradient of an advantage rate function and establish a martingale characterization
for both the value function and the advantage rate. These theoretical results provide tractable estimators for deterministic policy gradients in continuous-time RL. Building on this foundation, we propose a model-free continuous-time Deep Deterministic Policy Gradient (CT-DDPG) algorithm that enables stable learning for general reinforcement learning problems with continuous time-and-state. Numerical experiments show that CT-DDPG achieves superior stability and faster convergence compared to existing stochastic-policy methods, across a wide range of learning tasks with varying time discretizations and noise levels.
报告人简介: Dr. Xin Guo holds the Coleman Fung Chair professorship and chairs the IEOR department at UC Berkeley. She previously held positions at Cornell (2003-2006) and IBM research (1999-2003). She is a well-recognized and influential scholar whose research spans stochastic processes, control and games, machine learning, and mathematical finance. She has been on the editorial boards of a number of leading journals, including Operations Research, Mathematics of Operations Research, SIAM Control and Optimization, and Mathematical Finance. Her work connects rigorous mathematical methods with important applications in
finance, data science, biology, and healthcare. Notably, her work has been adopted by industry, with hundreds of millions of dollars in cost savings. She has also laid the mathematical foundation for some early- cancer-detection methodologies, approved by FDA.
邀请人：梁宗霞
