BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Learning generative models of images by factoring appearance and s
 hape - Nicolas Heess
DTSTART:20110329T083000Z
DTEND:20110329T103000Z
UID:TALK30437@talks.cam.ac.uk
CONTACT:Microsoft Research Cambridge Talks Admins
DESCRIPTION:Rich prior knowledge of the visual world is crucial for many v
 ision tasks\, and much effort has been devoted to formalizing such knowled
 ge for use in computer vision systems and elsewhere. Generative\, probabil
 istic models are an appealing framework for this purpose: they allow us to
  reason about uncertainty\, and importantly\, they are particularly amenab
 le to unsupervised learning.  Despite considerable efforts however\, a com
 prehensive generative model that can represent image structure at differen
 t levels of abstraction and scale and that still allows for efficient infe
 rence and learning remains largely elusive. \n\nIn my talk I will discuss 
 some steps towards this long-term goal. One hallmark of natural images is 
 the variability of visual characteristics across different image regions a
 nd the presence of sharp boundaries between regions which arise from objec
 ts occluding each other. Many generative models of generic natural images 
 have difficulties representing this type of structure and I will present a
  model that addresses this problem. It builds on concepts from the compute
 r vision literature such as the layered representation of images and combi
 nes them with ideas from ‘deep’\, unsupervised learning. I will first 
 describe the basic building block of the model\, the Masked Restricted Bol
 tzmann Machine\, which allows occlusion boundaries to be modeled by factor
 ing out the appearance of an image region from its shape. This model also 
 has a natural extension to images of realistic size: the Field of Masked R
 BMs models an image in terms of a large number of independent small and pa
 rtially overlapping ‘objects’\, each of which has an associated shape 
 and appearance. Finally\, I will discuss where this leaves us in the quest
  for a comprehensive representation of visual structure. I will give an ou
 tlook of how the Field of Masked RBMs naturally gives rise to a compositio
 nal\, hierarchical framework for modeling images at different scales and l
 evels of abstraction\, and I will talk about some of the challenges ahead.
 \n\nJoint work with Nicolas Le Roux\, John Winn\, and Jamie Shotton\n
LOCATION:Small lecture theatre\, Microsoft Research Ltd\, 7 J J Thomson Av
 enue (Off Madingley Road)\, Cambridge
END:VEVENT
END:VCALENDAR