Once a month during the academic year, the statistics faculty select a paper for our students to read and discuss. Papers are selected based on their impact or historical value, or because they contain useful techniques or results.
Diffusion Models
Prof. Xingche Guo
Friday, October 23 Austin 326 Noon to 1:00 PM
AUST 326
Webex Link: https://uconn-cmr.webex.com/uconn-cmr/j.php?MTID=mfaa46d753970e44c30bd16b45d4063e8
Abstract: Recent advances in generative modeling have enabled remarkable progress in text-to-image and text-to-video generation. Diffusion models are central to many of these developments, offering a powerful framework for generating complex, high-dimensional data by learning to reverse a gradual noise-injection process. This talk introduces the essential ideas and statistical foundations of diffusion models, emphasizing the connections among several complementary perspectives. The first half develops the basic diffusion framework and explains how its training objective arises from variational inference. The second half explores diffusion models through score-based modeling and stochastic differential equations, showing how these viewpoints connect and provide further insight into training and sampling. Time permitting, the talk will also introduce flow matching and discuss its relationship to diffusion-based generation.
Pizza will be served – RSVP is required