Optimal perception under uncertainty: Testing Bayesian models of vision
11 / 2009 - 11 / 2011
Human perception depends on uncertain inputs to infer the state of the world. Uncertainty comes from ambiguities in the world as well as neuronal noise, for example when estimating the distance to the next car on a foggy road or when trying to understand someone?s words at a noisy cocktail party. It is crucial that the brain takes uncertainty into account when making perceptual decisions, so that more reliable stimuli can be given more weight and performance is optimized. This idea is formalised in the theory of Bayesian inference. Recent behavioural studies suggest that the brain indeed processes sensory information in a Bayes-optimal way. We will reconsider a number of open problems in vision research from a Bayesian perspective. We will focus on two major questions: 1) To what extent can human visual perception be modelled as a Bayesian inference process? 2) What neural mechanisms underlie these perceptual decisions? These questions will be approached using a synthesis of behavioural experiments and computational methods, in which theoretical predictions are directly tested experimentally. Specifically, we will study visual search and visual crowding tasks. The results of this project will significantly contribute to our understanding of how the visual system processes uncertain information.