24.05.2024
When: On May 27, 2024 from 11:15 a.m. to 12:45 p.m.
Where: Joseph-von-Fraunhofer Straße 25, room 303 + Zoom
Open for: All who are interested
About the speaker:
Stephan Mandt is an Associate Professor of Computer Science and Statistics at the University of California, Irvine. His research topics include deep generative modeling, uncertainty quantification, neural data compression, and AI for science. He received his doctorate in theoretical physics from the University of Cologne. Other positions held by Stephan Mandt include Disney Research, Princeton and Columbia University.
Abstract:
Denoising diffusion models have led to a series of breakthroughs in image and video generation. In this talk, I will explore the deep connections between diffusion models and physics. Rooted in non-equilibrium thermodynamics, these models enable a variety of extensions by lifting diffusions into augmented spaces, encompassing position, momentum, and potentially additional variables. Augmented diffusions provide a “complete recipe” for constructing invertible diffusion processes, as well as new samplers that significantly enhance their inference efficiency. Additionally, thermodynamic processes offer a natural playground for generative AI. I will demonstrate how video diffusion models can effectively downscale precipitation patterns to finer scales, precisely capturing extreme event statistics and local geographical patterns.