BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Optimal weighted nearest neighbour classifiers - Dr Richard Samwor
 th
DTSTART:20110217T140000Z
DTEND:20110217T153000Z
UID:TALK29122@talks.cam.ac.uk
CONTACT:Shakir Mohamed
DESCRIPTION:Classifiers based on nearest neighbours are perhaps the simple
 st\nand most intuitively appealing of all nonparametric classifiers. Argua
 bly\nthe most obvious defect with the $k$-nearest neighbour classifier is 
 that\nit places equal weight on the class labels of each of the $k$ neares
 t\nneighbours to the point being classified.  Intuitively\, one would expe
 ct\nimprovements in terms of the misclassification rate to be possible by\
 nputting decreasing weights on the class labels of the successively more\n
 distant neighbours.  In this talk\, we determine the asymptotically optima
 l\nweighting scheme\, and quantify the benefits attainable.  Notably\, the
 \nimprovements depend only on the dimension of the feature vectors\, and n
 ot\non the underlying population densities.  We also show that the bagged\
 nnearest neighbour classifier falls within our framework\, and compare it\
 nwith the optimal weighted nearest neighbour classifier.\n\nThe talk will 
 be based on the following paper\nhttp://arxiv.org/abs/1101.5783
LOCATION:Engineering Department\, CBL Room 438
END:VEVENT
END:VCALENDAR