Title: “Bayesian Inferences with Population Codes”

Alexandre Pouget

University of Rochester

Abstract:

The brain is constantly performing computations over variables whose values cannot be known with certainty. A natural way to deal with this uncertainty is to perform statistical inferences over the variables, which requires a representation of the variables and their corresponding probability distributions. In the case of Gaussian distributions, we will show how the dynamics of basis function networks with multidimensional attractors can be used to perform close to optimal statistical inferences. This approach can be applied to a variety of problems including function approximation, multicue integration, temporal integration, motor control and decision theory. In the particular context of multisensory integration, our architecture predicts the existence of neurons with a particular type of receptive field, known as partially shifting receptive field, which has been recently reported in the cortex.