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:Moment matching for latent variable models: from I
 CA to LDA and CCA - Professor Francis Bach (INRIA\
 , ENS)
DTSTART;TZID=Europe/London:20161006T110000
DTEND;TZID=Europe/London:20161006T120000
UID:TALK68096AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/68096
DESCRIPTION:Moment matching is a traditional alternative to ma
 ximum likelihood for parameter estimation in proba
 bilistic models. For certain latent variable model
 s\, this has recently led to parameter estimation 
 algorithms with theoretical guarantees. While inde
 pendent component analysis (ICA) was the first sem
 i-parametric model to be considered twenty years a
 go\, this has been recently extended to latent Dir
 ichlet Allocation (LDA)\, which is a parametric mo
 del for discrete data. In this talk I will present
  (a) a semi-parametric extension of LDA which\, be
 yond making fewer modelling assumptions\, leads to
  simpler estimation through moment matching techni
 ques\, and (b) an extension to multi-view models s
 uch as canonical correlation analysis (CCA). (Join
 t work with Anastasia Podosinnikova and Simon Laco
 ste-Julien).
LOCATION:CBL Room BE-438
CONTACT:Zoubin Ghahramani
END:VEVENT
END:VCALENDAR