BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Wasserstein GANs Work Because They Fail (to Approximate the Wasser
 stein Distance) - Jan Stanczuk (University of Cambridge)
DTSTART:20210602T130000Z
DTEND:20210602T140000Z
UID:TALK159964@talks.cam.ac.uk
CONTACT:Neil Deo
DESCRIPTION:Wasserstein GANs are based on the idea of minimising the Wasse
 rstein distance between a real and a generated distribution. We provide an
  in-depth mathematical analysis of differences between the theoretical set
 up and the reality of training Wasserstein GANs. In this work\, we gather 
 both theoretical and empirical evidence that the WGAN loss is not a meanin
 gful approximation of the Wasserstein distance. Moreover\, we argue that t
 he Wasserstein distance is not even a desirable loss function for deep gen
 erative models\, and conclude that the success of Wasserstein GANs can in 
 truth be attributed to a failure to approximate the Wasserstein distance.\
 n\nThis talk is based off of the following joint work: https://arxiv.org/p
 df/2103.01678.pdf
LOCATION:https://maths-cam-ac-uk.zoom.us/j/95531783868?pwd=U3pPbmYxTXZYRVZ
 MWFBVTkVnWmUvZz09
END:VEVENT
END:VCALENDAR