BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Stochastic Optimization for Wasserstein Estimators - Marin Ballu (
 University of Cambridge)
DTSTART:20210609T130000Z
DTEND:20210609T140000Z
UID:TALK159952@talks.cam.ac.uk
CONTACT:Neil Deo
DESCRIPTION:Optimal transport is a foundational problem in\noptimization\,
  that allows to compare probability\ndistributions while taking into accou
 nt geometric\naspects. Its optimal objective value\, the Wasserstein dista
 nce\, provides an important loss between\ndistributions that has been used
  in many applications throughout machine learning and statistics.\nRecent 
 algorithmic progress on this problem and\nits regularized versions have ma
 de these tools increasingly popular. However\, existing techniques (pre-20
 20)\nrequire solving an optimization problem to obtain a single gradient o
 f the loss\, thus slowing\ndown first-order methods to minimize the sum of
 \nlosses\, that require many such gradient computations. In this talk\, I 
 will introduce an algorithm\nto solve a regularized version of this proble
 m\nof Wasserstein estimators\, with a time per step\nwhich is sublinear in
  the natural dimensions of\nthe problem.
LOCATION:https://maths-cam-ac-uk.zoom.us/j/95531783868?pwd=U3pPbmYxTXZYRVZ
 MWFBVTkVnWmUvZz09
END:VEVENT
END:VCALENDAR