Continuous restricted Boltzmann machine with an implementable training algorithm

Abstract

The authors introduce a continuous stochastic generative model that can model continuous data, with a simple and reliable training algorithm. The architecture is a continuous restricted Boltzmann machine, with one step of Gibbs sampling, to minimise contrastive divergence, replacing a time-consuming relaxation search. With a small approximation, the training algorithm requires only addition and multiplication and is thus computationally inexpensive in both software and hardware. The capabilities of the model are demonstrated and explored with both artificial and real data.