Beyond Generation: Flow Matching for Anomaly Detection and Reinforcement Learning

Generative models such as diffusion and flow matching have transformed how we learn complex distributions, while anomaly detection and reinforcement learning offers principled tools for sequential prediction and decision making. This talk explores how the two fields can strengthen each other, with an emphasis on the how to learn and exploit the full distribution information learned by the generative model. I first show how generative models give reinforcement learning a richer view of uncertainty, learning entire critics distributions from offline data, and discuss the statistical questions this raises. I then show how ideas from reinforcement learning and control let us use a trained flow model beyond generation, as a principled tool for anomaly detection.

Bio
Hao Yan is an Associate Professor in the School of Computing and Augmented Intelligence at Arizona State University. He received his Ph.D. in Industrial Engineering from the Georgia Institute of Technology. His research lies at the intersection of statistical learning, generative modeling, and reinforcement learning, with a focus on anomaly detection, prognostics, and uncertainty-aware decision making in complex engineered systems. Applications include manufacturing quality, reliability, and system health monitoring. His work has received more than 20 best paper recognitions, including IISE Transactions, IEEE Transactions on Automation Science and Engineering, and INFORMS.