A neural network with a biologically possible architecture can implement Bayesian estimation and reproduce Piéron's law

Abstract

Bayesian estimation theory has been expected to explain how the brain deals with uncertainty. Several previous studies have implied that cortical network models could implement Bayesian computation. However, the feasibility of the required computational procedures is still unclear under the physiological and anatomical constraints of neural systems. Here, we propose a neural network model that implements the algorithm in a biologically realizable manner, incorporating discrete choice theory. Our model successfully demonstrates an orientation discrimination task with significantly noisy visual images and the relation between the stimulus intensity and the reaction time known as Piéron’s law.