BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Dissipation and Optimal Transport in Discrete and Continuous Flow-
 based Models  - Grant M. Rotskoff (U Stanford)
DTSTART:20250513T120000Z
DTEND:20250513T130000Z
UID:TALK229843@talks.cam.ac.uk
CONTACT:Sarah Loos
DESCRIPTION: In this talk\, I will discuss connections between control pro
 blems that arise in nonequilibrium statistical physics and generative mode
 ls to illustrate how tools built for low dissipation control and rare even
 t sampling can be leveraged to improve sample efficiency in flow-based gen
 erative models. Employing these notions\, I will outline a strategy for pa
 rameterizing discrete diffusion models using tensor networks\, which impro
 ves MCMC in some simple models from statistical mechanics. Going beyond th
 is approach\, I will discuss "mixed-resolution" discrete and continuous mo
 dels we have been developing and how to build statistically controlled sam
 pling schemes for these models.
LOCATION:Center for Mathematical Sciences\, Lecture room MR4
END:VEVENT
END:VCALENDAR