Cortical Belief Networks

An alternative hypothesis is proposed, that neural populations represent, and effectively compute, probabilities, and under this hypothesis, population activities can contain additional information about such things as multiple values of, or uncertainty about, the quantity.

Abstract

Most theoretical and empirical studies of cortical population codes make the assumption that underlying neuronal activities is a unique and unambiguous value of an encoded quantity. We propose an alternative hypothesis, that neural populations represent, and effectively compute, probabilities. Under this hypothesis, population activities can contain additional information about such things as multiple values of, or uncertainty about, the quantity. We discuss methods for recovering this extra information, and show how this approach bears on psychophysical and neurophysiological studies. A natural extension of this probabilistic interpretation hypothesis casts interacting populations as a belief network, a structure which permits the analysis of information propagation from one population to another. This novel framework for population codes opens up new avenues for studying a diverse set of problems, including cue combination, decision-making, and visual attention.