BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Deep learning for vision: a case study for visual textures\,  and 
 some thoughts on a general framework - Prof. Chris Williams ( School of In
 formatics\, University of Edinburgh)
DTSTART:20120808T130000Z
DTEND:20120808T140000Z
UID:TALK39195@talks.cam.ac.uk
CONTACT:Zoubin Ghahramani
DESCRIPTION:I will argue that it is highly desirable to frame scene unders
 tanding\nin terms of hierarchical probabilistic models. These allow top-do
 wn\nand bottom-up flows of information to take place\, in order to provide
 \na scene interpretation. Encoded within such a model would be knowledge\n
 at various levels\, e.g. lower-level models of regions and boundaries\,\na
 nd at a higher level the shape and appearance of object classes\, and\nthe
 ir contextual relationships. I will further argue that such models\nshould
  be learned in a largely unsupervised fashion from image\ndata. Hinton's "
 deep learning" agenda is attractive here in that it\nprovides an upgrade p
 ath from lower-level to higher-level\nregularities.\n\nWe assess the gener
 ative power of the mPoT-model of Ranzato et al\n(2010) with tiled-convolut
 ional weight sharing as a model for visual\ntextures by specifically train
 ing on this task\, evaluating model\nperformance on texture synthesis and 
 inpainting tasks using\nquantitative metrics. We also analyze the relative
  importance of the\nmean and covariance parts of the mPoT model by compari
 ng its\nperformance to those of its subcomponents.  Our results suggest th
 at\nwhile state-of-the-art or better performance can be achieved using the
 \nmPoT\, similar performance can be achieved with the mean-only model. \nW
 e then extend this model to handle multiple textures\, using a shared\nset
  of weights but texture-specific hidden unit biases. We show\ncomparable p
 erformance of the multiple texture model to individually\ntrained texture 
 models\, providing state-of-the-art results in the\ntexture synthesis and 
 inpainting tasks.\n\nJoint work with Jyri Kivinen\, Ali Eslami\, Nicolas H
 eess.
LOCATION:Engineering Department\, CBL Room BE-438
END:VEVENT
END:VCALENDAR