BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Function Space Diffusion for Video Modeling - Nikola Kovachki (Non
 e / Other)
DTSTART:20240718T133000Z
DTEND:20240718T143000Z
UID:TALK219055@talks.cam.ac.uk
DESCRIPTION:We present a generalization of score-based diffusion models to
  function space by perturbing functional data via a Gaussian process at mu
 ltiple scales. We obtain an appropriate notion of score by defining densit
 ies with respect to Guassian measures and generalize denoising score match
 ing. We then define the generative process by integrating a function-value
 d Langevin dynamic. We show that the corresponding discretized algorithm g
 enerates samples at a fixed cost that is independent of the data discretiz
 ation. As an application for such a model\, we formulate video generation 
 as a sequence of joint inpainting and interpolation problems defined by fr
 ame deformations. We train an image diffusion model using Gaussian process
  inputs and use it to solve the video generation problem by enforcing equi
 variance with respect to frame deformations. Our results are state-of-the-
 art for video generation using models trained only on image data.
LOCATION:External
END:VEVENT
END:VCALENDAR