Neural Sampling by Irregular Gating Inhibition of Spiking Neurons and Attractor Networks

TL;DR

This paper introduces a neural network model based on irregular gating inhibition that implements MCMC sampling, offering insights into neural inference, stochastic synaptic weights, and potential applications in biological and neuromorphic systems.

Contribution

It presents a novel neural network model using irregular gating inhibition that analytically implements MCMC sampling for neural inference.

Findings

Model successfully implements MCMC sampling in neural networks.

Irregular gating inhibition can produce stochastic synaptic weights.

Applicable to both neural attractor networks and spiking neuron models.

Abstract

A long tradition in theoretical neuroscience casts sensory processing in the brain as the process of inferring the maximally consistent interpretations of imperfect sensory input. Recently it has been shown that Gamma-band inhibition can enable neural attractor networks to approximately carry out such a sampling mechanism. In this paper we propose a novel neural network model based on irregular gating inhibition, show analytically how it implements a Monte-Carlo Markov Chain (MCMC) sampler, and describe how it can be used to model networks of both neural attractors as well as of single spiking neurons. Finally we show how this model applied to spiking neurons gives rise to a new putative mechanism that could be used to implement stochastic synaptic weights in biological neural networks and in neuromorphic hardware.