Learning Priors for Bayesian Computations - Rehabilitation Institute of Chicago

Skip to Content

Fri, March 26

Speaker: Max Berniker, PhD

Title: Learning Priors for Bayesian Computations

Abstract: Numerous studies have found that subjects integrate their observations with their previous experience (priors) in a way that is close to the statistical optimum.  However, little is known about the way the nervous system acquires or learns priors.  Here we present results from experiments where the underlying distribution of target locations in an estimation task was switched, manipulating the prior subjects should use.  Our experimental design allowed us to measure the evolving prior while subjects learn.  We present the results from a group of unimpaired subjects and preliminary results from schizophrenic patients.  We found that unimpaired individuals rapidly learn the mean of a new prior while the variance is learned more slowly and with a variable learning rate.  In addition, we found that a model of optimal inference could predict the time course of the observed learning while offering an intuitive explanation for the findings. In contrast, we find that schizophrenic patients show deficits in learning a prior.  These preliminary results offer an explanation for how the nervous system continuously updates it priors to enable efficient behavior, and how this process, when faulty, might help explain some positive symptoms of schizophrenia.