probabilistic models of human visual perception

From: George Dvorsky (
Date: Tue Apr 04 2006 - 19:41:40 MDT

extracting prior expectations and noise characteristics from
psychophysical experiments

sensory percepts result from a fusion of our current sensory input and
our internal expectations. more formally, bayesian inference specifies
how an observer can fuse uncertain sensory measurements with prior
expectations to compute an optimal estimate of a quantity in the world.
in this work we show that this theory provides an excellent description
of human speed perception. we are also able to "reverse-engineer" the
sensory uncertainty and prior expectations of human observers by
experimentally gathering their responses to moving stimuli.
collaborators: alan stocker and eero simoncelli
papers: NIPS 2004 | nature neuroscience 2006
abstracts: COSYNE 2004 | VSS2005 (journal of vision (5):8, abstract 928)
comments: news and views | scienceNOW | faculty1000

physiological implementations of bayesian models of perception

if humans are indeed bayes optimal observers then we would like to know
how such computational framework is implemented in the brain. we suggest
two alternative mechanisms for the example of visual motion estimation.
we argue that the brain does not need an explicit representation of
probability in populations of neurons in order to behave optimally.
collaborators: alan stocker and eero simoncelli
abstracts: COSYNE 2005

adaptation within a bayesian framework of perception

it is a fundamental property of biological sensory systems that they
adapt their sensory behavior in response to the context of recent
sensory experiences. is important to understand how adaptation can be
incorporated in a bayesian estimation framework. here we show that
adaptation is unlikely to impose a change in the applied prior
distribution. rather, the typical adaptation effects of repulsion and
increased discrimination thresholds seem consistent with an reduction in
the observation noise of the bayesian observer.
collaborators: alan stocker and eero simoncelli
paper: NIPS 2005
abstracts: COSYNE 2006

This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:56 MDT