Joint Training of Deep Boltzmann Machines

TL;DR

This paper presents a novel joint training method for deep Boltzmann machines that improves upon previous greedy layer-wise approaches, especially for classification tasks.

Contribution

It introduces a new joint training technique for deep Boltzmann machines that enhances training efficiency and classification performance.

Findings

Joint training outperforms greedy layer-wise methods.

Improved classification accuracy on benchmark datasets.

Reduces training complexity for deep Boltzmann machines.

Abstract

We introduce a new method for training deep Boltzmann machines jointly. Prior methods require an initial learning pass that trains the deep Boltzmann machine greedily, one layer at a time, or do not perform well on classifi- cation tasks.