Fluid Dynamic Models in Machine Learning

National AI Research Lab (NAIRL) hosted an invited talk by Daniel Dongyuel Lee, Tisch University Professor in Electrical and Computer Engineering at Cornell Tech, on November 12, 2024, at the Kim Dongmyung Lecture Hall on KAIST Seoul Campus.

Under the theme “Fluid Dynamic Models in Machine Learning,” Professor Lee explored how fluid dynamic models play an important role in understanding the dynamics of recent machine learning models, including reverse diffusion models and feature learning via stochastic gradient descent. He described connections between fluid flow models and probabilistic models, highlighting the role of conservation laws in both contexts. In particular, he showed how the Fokker-Planck equation provides a framework for describing how velocity flow fields and diffusive processes influence the spatial and temporal evolution of probability distributions, and discussed how these concepts apply to modeling neural weights and kernel integral operators in neural networks during feature learning.

Professor Lee recently served as Global Head of AI for Samsung Research. He received his B.A. summa cum laude in Physics from Harvard University and his Ph.D. in Condensed Matter Physics from MIT, and was a researcher at Bell Labs in the Theoretical Physics and Biological Computation departments. A Fellow of the IEEE and AAAI, he has received the NSF CAREER award and the Lindback award for distinguished teaching, and has organized the US-Japan National Academy of Engineering Frontiers of Engineering symposium and the Neural Information Processing Systems (NeurIPS) conference. His group focuses on understanding general computational principles in biological systems and applying that knowledge to build autonomous systems. NAIRL will continue to connect Korea’s research community with leading AI experts from around the world through its invited talk series.