BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Joint Gaussian Process-Density Mixtures - Ole Winther\, DTU / Infe
 rence Group
DTSTART:20050511T130000Z
DTEND:20050511T140000Z
UID:TALK4324@talks.cam.ac.uk
CONTACT:Phil Cowans
DESCRIPTION:Gaussian Processes (GPs) provide a natural framework for Bayes
 ian kernel\nmethods. This talk will be about some work in progress on comb
 ining GPs\nwith density estimation in a mixture model. The motivations are
 : using\nkernels tuned individually to each mixture component gives a more
  flexible\ninput-output model\, unlabelled data can be used in a semi-supe
 rvised\nsetting and the computational complexity can be reduced because on
 ly\nexamples belonging to the same mixture component need to be included i
 n\nthe kernel matrix for that mixture component. I will illustrate the ide
 a\nwith a regression example using a coarse two-stage approximation: densi
 ty\nestimation followed by weighted GP predictions. A more principled\nvar
 iational Bayes treatment of the joint estimation problem shows how a\nlow 
 complexity solution can be obtained.
LOCATION:Ryle Seminar Room\, Cavendish Laboratory
END:VEVENT
END:VCALENDAR