BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Flow matching\, stochastic interpolants and everything in between 
 - Emile Mathieu\, Tor Fjelde and Vincent Dutordoir (Cambridge MLG)
DTSTART:20231129T110000Z
DTEND:20231129T123000Z
UID:TALK209146@talks.cam.ac.uk
CONTACT:Isaac Reid
DESCRIPTION:Flow matching is the latest development in deep generative mod
 elling and has already been applied to numerous tasks including protein de
 sign\, image inpainting etc. Flow matching brought continuous normalising 
 flows to the front stage\, showing that they can be trained 'simulation-fr
 ee'\, as in without solving an ODE at each training step. These are closel
 y related to diffusion models\, yet yielding noiseless trajectories at sam
 pling. We will introduce relevant core ideas\, and discuss the advantage a
 nd inconvenience of noisy trajectories.\n\nSuggested reading:\n1. Lipman\,
  Chen & Ben-Hamu et al. (2022) Flow Matching for Generative Modeling. URL:
  http://arxiv.org/abs/2210.02747.\n2. Tong\, Malkin & Huguet et al. (2023)
  Improving and Generalizing Flow-Based Generative Models With Minibatch Op
 timal Transport. URL: http://arxiv.org/abs/2302.00482.\n3. Albergo\, Boffi
  & Vanden-Eijnden (2023) Stochastic Interpolants: a Unifying Framework for
  Flows and Diffusions. URL: http://arxiv.org/abs/2303.08797
LOCATION:Cambridge University Engineering Department\, CBL Seminar room BE
 4-38.
END:VEVENT
END:VCALENDAR