Training Restricted Boltzmann Machine by Perturbation

TL;DR

This paper introduces Perturb and Descend, a novel training method for RBMs that uses perturbation-based sampling and energy landscape descent, leading to efficient learning and regularization effects.

Contribution

The paper proposes a new perturbation-based training algorithm for RBMs that improves efficiency and model robustness compared to existing methods.

Findings

Perturb and Descend effectively trains RBMs using linear calculations.

The method produces sparse hidden layer activations, enhancing feature robustness.

Perturbation magnitude acts as a regularizer, improving model generalization.

Abstract

A new approach to maximum likelihood learning of discrete graphical models and RBM in particular is introduced. Our method, Perturb and Descend (PD) is inspired by two ideas (I) perturb and MAP method for sampling (II) learning by Contrastive Divergence minimization. In contrast to perturb and MAP, PD leverages training data to learn the models that do not allow efficient MAP estimation. During the learning, to produce a sample from the current model, we start from a training data and descend in the energy landscape of the "perturbed model", for a fixed number of steps, or until a local optima is reached. For RBM, this involves linear calculations and thresholding which can be very fast. Furthermore we show that the amount of perturbation is closely related to the temperature parameter and it can regularize the model by producing robust features resulting in sparse hidden layer activation.