BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Factored Shapes and Appearances for Parts-based Object Understandi
 ng AND Transformation Equivariant Boltzmann Machines - Chris Williams\, Sc
 hool of Informatics\, University of Edinburgh
DTSTART:20110927T100000Z
DTEND:20110927T110000Z
UID:TALK33320@talks.cam.ac.uk
CONTACT:Carl Edward Rasmussen
DESCRIPTION:Two short talks:\n\n# Factored Shapes and Appearances for Part
 s-based Object Understanding\n\nWe present a novel generative framework fo
 r learning parts-based\nrepresentations of object classes. Our model\, Fac
 tored Shapes and\nAppearances (FSA)\, employs a highly factored representa
 tion to reason about appearance and shape variability across datasets of i
 mages. We propose Markov Chain Monte Carlo sampling schemes for efficient 
 inference and learning\, and evaluate the model on a number of datasets. H
 ere we consider datasets that exhibit large amounts of variability\, both 
 in the shapes of objects in the scene\, and in their appearances. We show 
 that the FSA model extracts meaningful parts from training data\, and that
  its parameters and representation can be used to perform a range of tasks
 \, including object parsing\, segmentation and fine-grained categorisation
 .\n\nJoint work with Ali Eslami\n\n# Transformation Equivariant Boltzmann 
 Machines\n\nWe develop a novel modeling framework for Boltzmann machines\,
 \naugmenting each hidden unit with a latent transformation assignment vari
 able which describes the selection of the transformed view of the canonica
 l connection weights associated with the unit. This enables the inferences
  of the model to transform in response to transformed input data in a stab
 le and predictable way\, and avoids learning multiple features differing o
 nly with respect to the set of transformations. Extending prior work on tr
 anslation equivariant (convolutional) models\, we develop translation and 
 rotation equivariant restricted Boltzmann machines (RBMs) and deep belief 
 nets (DBNs)\, and demonstrate their effectiveness in learning frequently o
 ccurring statistical structure from artificial and natural images.\n\nJoin
 t work with  Jyri Kivinen\n
LOCATION:Engineering Department\, CBL Room 438
END:VEVENT
END:VCALENDAR