BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Clustering by linear programming\, convex optimization and belief 
 propagation - Brendan Frey\, University of Toronto
DTSTART:20080916T130000Z
DTEND:20080916T140000Z
UID:TALK13533@talks.cam.ac.uk
CONTACT:David MacKay
DESCRIPTION:\nA popular approach to clustering is to identify a small set 
 of data points\ncalled exemplars\, and associate every other data point wi
 th an exemplar.\nThe goal is to maximize the sum of similarities between d
 ata points and\ntheir exemplars. This method can be used to cluster vector
 -space data\, but\ncan also be applied to non-vector and even non-metric d
 ata\, since all that\nis needed is a set of similarities between pairs of 
 data points. In fact\,\ndata points and exemplars can come from different 
 spaces\, eg\, the data\npoints could be disaster victims while the exempla
 rs are potential food\nrepositories\, or the data points could be regions 
 of space to be imaged\nwhile the exemplars are potential telescopes. In th
 is talk\, I'll review\nthe state-of-the-art in algorithms for exemplar-bas
 ed clustering\,\nincluding recently-proposed ones based on convex optimiza
 tion\, loopy\nbelief propagation and linear programming. I'll also present
  benchmarks\nfor these methods.\n\n
LOCATION:TCM Seminar Room\, Cavendish Laboratory\, Department of Physics
END:VEVENT
END:VCALENDAR