BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Not so naive Bayesian classification - Prof. Geoff Webb (Monash Un
 iv)
DTSTART:20111012T100000Z
DTEND:20111012T110000Z
UID:TALK33042@talks.cam.ac.uk
CONTACT:Zoubin Ghahramani
DESCRIPTION:Machine learning is classically conceived as search through a 
 hypothesis space for a hypothesis that best fits the training data. In con
 trast\, naive Bayes performs no search\, extrapolating an estimate of a hi
 gh-order conditional probability by composition from lower-order condition
 al probabilities. In this talk I show how this searchless approach can be 
 generalised\, creating a family of learners that provide a principled meth
 od for controlling the bias/variance trade-off. At one extreme very low va
 riance can be achieved as appropriate for small data.  Bias can be decreas
 ed with larger data \nin a manner that ensure Bayes optimal asymptotic err
 or.  These algorithms havethe desirable properties of\n* training time tha
 t is linear with respect to training set size\,\n* learning rom a single p
 ass through the data\,\n* allowing incremental learning\,\n* supporting pa
 rallel and anytime classification\,\n* providing direct prediction of clas
 s probabilities\,\n* supporting direct handling of missing values\, and\n*
  robust handling of noise.\n\nDespite being generative\, they deliver clas
 sification accuracy competitive with state-of-the-art discriminative techn
 iques.\n
LOCATION:Engineering Department\, CBL Room 438
END:VEVENT
END:VCALENDAR