BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:The power of two samples in Generative Adversarial Networks (GAN).
DTSTART:20180109T110000
DTEND:20180109T120000
DTSTAMP:20260916T034316Z
UID:3dea319e3f6f0caa662bbfecc068a3b0aeca186f320c3682ed00b932
CATEGORIES:Conferences - Seminars
DESCRIPTION:Sewoong Oh\, UIUC\nWe bring the tools from Blackwell's seminal
  result in 1958 on comparing two stochastic experiments\, to shine new lig
 hts on a modern  applications of great interest: generative adversarial n
 etworks (GAN). Binary hypothesis testing is at the center of this applicat
 ion\, and we propose new data processing inequalities that allows us to di
 scover new algorithms\, provide sharper analyses\, and provide simpler pro
 ofs. This leads to a new framework to handle one of the major challenges i
 n GAN known as ``mode collapse''\; the lack of diversity in the samples ge
 nerated by the learned generators.\n\nThe hypothesis testing view of GAN a
 llows us to make a fundamental connection between our proposed idea of "pa
 cking" and mode collapse\, suggesting that packing is the right framework 
 to deal with mode collapse\, when training GANs. For this talk\, I will as
 sume no prior background on GAN.
LOCATION:INR113 http://plan.epfl.ch/?lang=fr&room=inr113
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR