BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Covariate Shift Adaptation: Supervised Learning When Training and 
 Test Inputs Have Different Distributions - Masashi Sugiyama (Tokyo Institu
 te of Technlogy)
DTSTART:20070913T120000Z
DTEND:20070913T133000Z
UID:TALK7807@talks.cam.ac.uk
CONTACT:Zoubin Ghahramani
DESCRIPTION:A common assumption in supervised learning is that the input p
 oints in\nthe training set follow the same probability distribution as the
  input\npoints in the test phase. However\, this assumption is not satisfi
 ed\,\nfor example\, when the outside of the training region is\nextrapolat
 ed. The situation where the training input points and test\ninput points f
 ollow different distributions while the conditional\ndistribution of outpu
 t values given input points is unchanged is\ncalled the covariate shift. U
 nder the covariate shift\, standard\ntechniques such as maximum likelihood
  estimation or cross validation\ndo not work as desired.  In this talk\, I
  will introduce covariate\nshift adaptation techniques which we developed 
 recently.\n\nReferences:\n\nSugiyama\, M.\, Krauledat\, M.\, & Mueller\, K
 .-R.\nCovariate shift adaptation by importance weighted cross validation.\
 nJournal of Machine Learning Research\, vol.8 (May)\, pp.985-1005\, 2007.\
 n"http://sugiyama-www.cs.titech.ac.jp/~sugi/2007/IWCV.pdf"\n\nSugiyama\, M
 .\, Nakajima\, S.\, Kashima\, H.\, von Buenau\, P. & Kawanabe\, M.\,\nDire
 ct importance estimation with model selection and\nits application to cova
 riate shift adaptation.\nTechnical Report TR07-0003\, Department of Comput
 er Science\,\nTokyo Institute of Technology\, Tokyo\, Japan\, 2007.\n"pdf 
 file url":http://www.cs.titech.ac.jp/~tr/reports/2007/TR07-0003.pdf\n
LOCATION:LR5\, Engineering\, Department of
END:VEVENT
END:VCALENDAR