Fractional Neural Sampling: A Theory of Spatiotemporal Probabilistic Computations in Neural Circuits

Abstract

A range of perceptual and cognitive processes have been characterized from the perspective of probabilistic representations and inference. To understand the neural circuit mechanism underlying these probabilistic computations, we develop a new theory based on complex spatiotemporal dynamics of neural population activity. We first implement and explore this theory in a biophysically realistic, spiking neural circuit. Population activity patterns emerging from the circuit capture the realistic variability or fluctuations of neural dynamics not only in time but also in space; such spatial fluctuations, however, are ignored in traditional models of neural computations. These activity patterns implement a new type of probabilistic computations that we call Fractional Neural Sampling (FNS). We further develop a normative model to reveal the algorithmic nature of FNS and its powerful computational advantages for representing multimodal distributions, a major challenge faced by existing theories. We show that FNS accounts for recent experimental observations of perceptual inference and makes experimentally testable predictions.