BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Principled Hybrids of Generative and Discriminative Models - Phili
 p Sterne (University of Cambridge)
DTSTART:20080204T111000Z
DTEND:20080204T124500Z
UID:TALK10656@talks.cam.ac.uk
CONTACT:Philip Sterne
DESCRIPTION:For this journal club we will look at "Principled Hybrids of G
 enerative and Discriminative Models":http://research.microsoft.com/~cmbish
 op/downloads/Bishop-CVPR-06.pdf by Bishop and Laserre (2006)\n\nA shortene
 d abstract:\nWhen labelled training data is plentiful\, discriminative tec
 hniques are widely used since they give excellent generalization performan
 ce.  However\, for large-scale applications such as object recognition\, h
 and labelling of data is expensive\, and there is much interest in semi-su
 pervised techniques based on generative models in which the training data 
 is unlabelled.   In an attempt to gain the benefit of both generative and 
 discriminative approaches\, heuristic procedure have been proposed [2\, 3]
  which interpolate between these two extremes.\n\nIn this paper we adopt a
  new perspective which says that\nthere is only one correct way to train a
  given model\, and\nthat a `discriminatively trained' generative model is 
 fundamentally a new model [7]. From this viewpoint\, generative and discri
 minative models correspond to specific choices for the prior over paramete
 rs. \n\nAnother related article is "Generative or Discriminative? Getting 
 the Best of Both Worlds":http://mi.eng.cam.ac.uk/~jal62/publis/isba06_hybr
 id_models.pdf\n\n*The journal club will be preceded by a 5-minute talk by 
 Ignas about his undergraduate project*
LOCATION:TCM Seminar Room\, Cavendish Laboratory\, Department of Physics
END:VEVENT
END:VCALENDAR