Reward-modulated synaptic plasticity for simple Bayesian decision

Gold and Shadlen have proposed that a simple quantity, logarithm likelihood ratio (logLR), can provide a neural currency for Bayesian inference of two alternative choice tasks. However, it remains unclear how our nervous system could acquire the capability to carry out this computation. In this paper we propose a learning rule operating on logLR. In particular we show that the experimentally supported type of reward-modulated synaptic rule in combination with a winner-take-all neural circuit can model the mainly experiment results by Yang and Shadlen.