WHO:	Ruben Moreno Bote , BCS Postdoc
TOPIC:	BCS Lunch Talk
ABSTRACT:	When an observer views a stimulus that allows two distinct interpretations, only one interpretation is perceived at any given time, and perception switches between the two in a stochastic manner. Well-known examples of this phenomenon, called perceptual bistability, are the Necker cube and the face-vase illusion. I will show experimental results that strongly suggest that the dynamics of perceptual bistability arises from a sampling process of an underlying probability distribution over the causes of the stimulus. Next, I will describe diffusion models embedded in double-well potentials, known to perform Langevin Monte-Carlo sampling, that generate the observed behaviors. The generality of this model will be illustrated with rate-based models and more biophysically realistic spiking neurons endowed with attractor dynamics. In conclusion, neuronal networks in the brain seem to sample probability distributions, at least during conditions of high uncertainty or ambiguity in the stimulus. Our results invite us to explore the hypothesis that the brain is an approximate inference machine that uses sampling algorithms throughout tasks and domains
WHEN:	11/18/2009 12:00:00 PM
WHERE:	Meliora 269

Events Homepage