Reconfigurable Stochastic neurons for Restricted Boltzmann Machine

Abstract

Neuromorphic implementation of Boltzmann Machine using a network of stochastic neurons can allow non-deterministic polynomial-time (NP) hard combinatorial optimization problems to be efficiently solved. Efficient implementation of such Boltzmann Machine with simulated annealing desires the statistical parameters of the stochastic neurons to be dynamically tunable, however, there has been limited research on stochastic semiconductor devices with controllable statistical distributions. Here, we demonstrate a fully digitized stochastic neurons based on reconfigurable stochastic model that can realize tunable stochastic dynamics in its output sampling characteristics. We emulate the Restricted Boltzmann Machine based on reconfigurable stochastic neurons and leverage the handwritten digits from the MNIST dataset to validate its recognition capabilities and investigate the accuracy of several distinct network parameters.