Our colloquium takes place on the first Tuesday of each month from 15:30 to 16:30, usually in room A709.
A renowned expert (being an excellent speaker as well) visits us for an afternoon and gives a panorama of one of her research areas. The talk is meant to be accessible to all members of the lab, including PhD students in analysis, game theory, probability and statistics. Ideally, it should start gently with an historical background on the problem and an overview of the main questions and applications, keeping a non technical style during at least the first half of the talk. Of course it is also nice to have a part with more mathematical details: the most appreciated colloquia were those in which the speaker succeeded to develop a nice technical idea or an elegant argument that everyone should know.
Food and drinks are served after the event, usually in Espace 7!
Date: Tuesday, December 2nd 2025 (15:30-16:30, room A709)
Title: Denoising diffusion models without diffusions
Abstract: Denoising diffusion models have enabled remarkable advances in generative modeling across various domains. These methods rely on a two-step process: first, sampling a noisy version of the data—an easier computational task—and then denoising it, either in a single step or through a sequential procedure. Both stages hinge on the same key component: the score function, which is closely tied to the optimal denoiser mapping noisy inputs back to clean data. In this talk, I will introduce an alternative perspective on denoising-based sampling that bypasses the need for continuous-time diffusion processes. This framework not only offers a fresh conceptual angle but also naturally extends to discrete settings, such as binary data. Joint work with Saeed Saremi and Ji-Won Park.