Linear-Nonlinear-Poisson Neuron Networks Perform Bayesian Inference On Boltzmann Machines

TL;DR

This paper demonstrates that biologically plausible Linear-Nonlinear-Poisson neuron networks can represent Boltzmann machines and perform Bayesian inference, bridging neuroscience models with machine learning algorithms.

Contribution

It establishes a detailed correspondence between LNP neuron networks and Boltzmann machines, enabling Bayesian inference in biologically plausible neural models.

Findings

LNP neuron networks can represent Boltzmann machines

Networks perform semi-stochastic Bayesian inference

Bridges neuroscience models with machine learning algorithms

Abstract

One conjecture in both deep learning and classical connectionist viewpoint is that the biological brain implements certain kinds of deep networks as its back-end. However, to our knowledge, a detailed correspondence has not yet been set up, which is important if we want to bridge between neuroscience and machine learning. Recent researches emphasized the biological plausibility of Linear-Nonlinear-Poisson (LNP) neuron model. We show that with neurally plausible settings, the whole network is capable of representing any Boltzmann machine and performing a semi-stochastic Bayesian inference algorithm lying between Gibbs sampling and variational inference.