BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Schrodinger Bridge Flow for Unpaired Data Translation - Valentin D
 e Bortoli (Google DeepMind Technologies Limited)
DTSTART:20240715T090000Z
DTEND:20240715T100000Z
UID:TALK219010@talks.cam.ac.uk
DESCRIPTION:Mass transport problems arise in many areas of machine learnin
 g whereby one wants to compute a map transporting one distribution to anot
 her. Generative modeling techniques like Generative Adversarial Networks (
 GANs) and Denoising Diffusion Models (DMMs) have been successfully adapted
  to solve such transport problems\, resulting in CycleGAN and Bridge Match
 ing respectively. However\, these methods do not approximate Optimal Trans
 port (OT) maps\, which are known to have desirable properties. Existing te
 chniques approximating OT maps for high-dimensional data-rich problems\, i
 ncluding DDMs-based Rectified Flow and Schrodinger bridge procedures\, req
 uire fully training a DDM-type model at each iteration\, or use mini-batch
  techniques which can introduce significant errors. We propose a novel alg
 orithm to compute the Schrodinger bridge\, a dynamic entropy-regularized v
 ersion of OT\, that eliminates the need to train multiple DDMs-like models
 . This algorithm corresponds to a discretization of a flow of path measure
 s\, referred to as the Schrodinger Bridge Flow\, whose only stationary poi
 nt is the Schrodinger bridge. We demonstrate the performance of our algori
 thm on a variety of unpaired data translation tasks.
LOCATION:External
END:VEVENT
END:VCALENDAR