Abstract: This talk examines the intersection of machine learning and optimal control through the development of scalable algorithms with rigorous theoretical guarantees. The first part presents learning-based frameworks for stochastic optimal control, with an emphasis on actor–critic methods that connect reinforcement learning and continuous-time control, together with their extensions to multi-agent mean field games. The second part illustrates how control-theoretic principles can advance modern machine learning by viewing generative models as controlled probability flows, leading an efficient and structure-preserving formulation, score-based neural ODEs and normalizing flows. Together, these results outline a unified framework in which machine learning and control mutually reinforce one another, offering new computational tools and analytical insights for scientific computing.

Short Bio: Mo Zhou is an Assistant Adjunct Professor in the Department of Mathematics at UCLA, working under the mentorship of Prof. Stanley Osher. He received his Ph.D. in Mathematics from Duke University in 2023, advised by Prof. Jianfeng Lu, and earned his bachelor’s degree in Mathematics from Tsinghua University in 2018. His research focuses on the interaction between optimal control and machine learning.

Tencent conference link: https://meeting.tencent.com/dm/5NtR0npBQRlx

Tencent conference number: 997-247-078