BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Handling obsolete information in classification: is there a one-si
 ze-fits-all strategy? - Christoforos Anagnostopoulos\, Imperial College Lo
 ndon
DTSTART:20121019T110000Z
DTEND:20121019T120000Z
UID:TALK39569@talks.cam.ac.uk
CONTACT:Ekaterina Kochmar
DESCRIPTION:Classifier performance is known to deteriorate over time\, as 
 a result of the\ntraining dataset becoming obsolete as time goes by. A par
 ticularly common\ninstance of this is the phenomenon of "population drift"
 \, whereby gradual\nchanges in the population characteristics have a cumul
 ative detrimental effect\non the accuracy of the classifier over time. In 
 online learning contexts\, where\nthe classifier is updated incrementally 
 as new information arrives\, this raises\nthe question of whether novel in
 formation should complement\, or altogether\nreplace older data. In this t
 alk\, we illustrate how such decisions hinge not\nonly on the precise natu
 re of the evolution exhibited by the data\, but also on\nthe methodology t
 hat underlies the classifier: different classifiers are\naffected in diffe
 rent ways by drift. This observation has been studied by\nvarious authors 
 in the streaming data literature\, and in this talk we review\nand extend 
 this work. We conclude by commenting on the suitability of\none-size-fits-
 all strategies for handling drift\, that monitor performance alone\nand ar
 e otherwise indifferent to the underlying methodology.
LOCATION:FW26\, Computer Laboratory
END:VEVENT
END:VCALENDAR