BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Learning transformational invariants from natural movies - Bruno O
 lshausen\, UC Berkeley
DTSTART:20090304T141500Z
DTEND:20090304T151500Z
UID:TALK16794@talks.cam.ac.uk
CONTACT:Rachel Fogg
DESCRIPTION:A key attribute of visual perception is the ability to extract
  invariances from visual input.  Here we focus on transformational invaria
 nts - i.e.\, the dynamical properties that are invariant to form or spatia
 l structure (e.g.\, motion).   We show that a hierarchical\,\nprobabilisti
 c model can learn to extract complex motion from movies of the natural env
 ironment.   The model consists of two hidden layers:  the first layer prod
 uces a sparse representation of the image that is expressed in terms of lo
 cal amplitude and phase variables. The second layer learns the higher-orde
 r structure among the time-varying phase variables. After training on natu
 ral movies\, the top layer units discover the structure of phase-shifts wi
 thin the first layer. We show that the top layer units encode transformati
 onal invariants: they are selective for the speed and direction of a movin
 g pattern\, but are invariant to its spatial structure (orientation/spatia
 l-frequency). The diversity of units in both the intermediate and top laye
 rs of the model provides a set of testable predictions for representations
  that might be found in V1 and MT.\nIn addition\, the model demonstrates h
 ow feedback from higher levels can influence representations at lower leve
 ls as a by-product of inference in a graphical model.\n\nJoint work with C
 harles Cadieu.\n
LOCATION:LR5\, Engineering\, Department of
END:VEVENT
END:VCALENDAR