BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Bayesian Learning for Visual Inference - Ollie Williams
DTSTART:20050323T150000Z
DTEND:20050323T160000Z
UID:TALK4304@talks.cam.ac.uk
CONTACT:Phil Cowans
DESCRIPTION:Every second\, gigabytes of data arrive at our eyes\, yet our 
 brain \neffortlessly translates this into concise descriptions of the worl
 d \nenabling us to perform everyday tasks. As information engineers\, our 
 \nresponsibility is to manage the overwhelming quantities of information \
 navailable to us and in my research I have taken inspiration from humans \
 nto learn the mappings between high-dimensional image data and a \nproblem
 -specific output space. Such mappings are learnt discriminatively \nfrom a
  set of labelled training data using the Bayesian rules of \ninference to 
 pragmatically account for uncertainty\, incorporate prior \nknowledge and 
 set parameter values. The benefits of learning mappings \n(as opposed to d
 efining a model of image generation) are efficiency and \nthe ability to g
 eneralize when images change in some previously unseen way.\nI will demons
 trate how this general concept has been used to create a \nsystem for trac
 king moving objects in video sequences and to create a \none-dimensional v
 isual interface that can be used to drive the Dasher \ntyping system.\nThi
 s is joint work with Andrew Blake (Microsoft Research) and Roberto \nCipol
 la (University of Cambridge).
LOCATION:Ryle Seminar Room\, Cavendish Laboratory
END:VEVENT
END:VCALENDAR