----------------------------------------------------------------------------------------------------

 

 

 

          Department of Systems Engineering and Engineering Management

 

 

 

                                  The Chinese University of Hong Kong

 

 

 

----------------------------------------------------------------------------------------------------

 

 

 

Date: Thursday, June 4, 2026, 2:30pm to 3:30pm HKT

 

Venue: ERB602, The Chinese University of Hong Kong

 

Title: Sobolev-Prox Algorithm for Continuous Time Reinforcement Learning

 

Speaker: Professor Wenlong Mou, University of Toronto

 

 

 

Abstract:

 

Model-free function approximation is the workhorse of modern reinforcement learning (RL). By solving projected Bellman fixed-point equations, these methods can learn the value functions and optimal policies efficiently without requiring a model of the environment. While function approximation is known to be hard for general discrete-time RL problems, recent work discovered that RL in continuous-time control problems exhibits remarkable properties that enable the design of provably efficient algorithms. In this talk, we introduce recent advances in optimization algorithms for continuous-time RL with function approximation. We present a new class of Sobolev-prox algorithms that leverage the function space structures of the Bellman operator from a controlled diffusion process. We prove non-asymptotic convergence guarantees for general function approximations. We will also discuss applications to guiding diffusion-based generative models.

 

 

 

Biography:

 

Wenlong Mou is an Assistant Professor in the Department of Statistical Sciences at the University of Toronto. He received his Ph.D. in Electrical Engineering and Computer Sciences from UC Berkeley in 2023. Before joining Berkeley, he earned a B.Sc. in Computer Science and a B.A. in Economics from Peking University. His research interests include reinforcement learning theory, post-training methods for deep generative models, and the interplay between reinforcement learning and continuous control. He is particularly interested in developing theory and algorithms that use reinforcement learning to control real-world systems such as fusion plasma. His work has been published in leading journals and conferences in machine learning, statistics, and applied mathematics. His research has been recognized by the INFORMS Applied Probability Society, where he was named a Best Student Paper finalist.

 

 

 

 

 

Everyone is welcome to attend the talk!

 

SEEM-5202 Website: http://seminar.se.cuhk.edu.hk