BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//talks.cam.ac.uk//v3//EN
BEGIN:VTIMEZONE
TZID:Europe/London
BEGIN:DAYLIGHT
TZOFFSETFROM:+0000
TZOFFSETTO:+0100
TZNAME:BST
DTSTART:19700329T010000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0100
TZOFFSETTO:+0000
TZNAME:GMT
DTSTART:19701025T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
CATEGORIES:Machine Learning @ CUED
SUMMARY:Bayesian Generative Adversarial Networks - Profess
 or Andrew Wilson\, Cornell University
DTSTART;TZID=Europe/London:20171213T110000
DTEND;TZID=Europe/London:20171213T120000
UID:TALK94141AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/94141
DESCRIPTION:Through an adversarial game\, generative adversari
 al networks (GANs) can implicitly learn rich distr
 ibutions over images\, audio\, and data which are 
 hard to model with an explicit likelihood.  I will
  present a practical Bayesian formulation for unsu
 pervised and semi-supervised learning with GANs.  
 Within this framework\, we use a stochastic gradie
 nt Hamiltonian Monte Carlo for marginalizing param
 eters.  The resulting approach can automatically d
 iscover complementary and interpretable generative
  hypotheses for collections of images\, without ad
  hoc interventions.  Moreover\, by exploring an ex
 pressive posterior over these hypotheses\, we show
  that it is possible to achieve state-of-the-art q
 uantitative results on image classification benchm
 arks\, even with less than 1% of the labelled trai
 ning data.  
LOCATION:CBL Seminar Room
CONTACT:Pat Wilson
END:VEVENT
END:VCALENDAR