NTR Webinar: Diffusion models with dynamically changing size

NTR organizes and hosts scientific webinars on neural networks and invites speakers from all over the world to present their recent work.

On April 5 Dmitry Vetrov, HSE, Moscow, Russia, presented a technical Zoom webinar on diffusion models, with dynamically changing size.

About the webinar: 

During the last two years, diffusion models have become a new SOTA in generative modeling.

This elegant concept brings together neural ODEs, score function estimators, variational Bayesian inference and stochastic differential equations. Its main drawback is the large computation time needed to generate an object.

Dmitry discussed the reasons for success of diffusion models and the ways to speed them up by reducing unnessesary computations at the beginning of reverse dynamics. The resulting model generalizes conventional diffusion models and hierarchical VAEs.

Slide deck link https://drive.google.com/file/d/1W8VCnfgCDcY1GN9F95SSnhtnJfSscCV-/view?usp=sharing

Moderator and contact: NTR CEO Nick Mikhailovsky: nickm@ntrlab.com.

Leave a Reply

Your email address will not be published. Required fields are marked *