Improved Classification Based on Deep Belief Networks

TL;DR

This paper introduces enhanced deep belief network models that integrate supervised learning into the generative pretraining process, leading to improved classification performance over traditional two-phase methods.

Contribution

The authors developed new supervised models based on deep belief networks that unify unsupervised and supervised objectives through modified loss functions and multi-level programming.

Findings

Models outperform traditional two-phase training methods

Modified loss functions improve generative and discriminative capabilities

Enhanced models effectively capture both learning objectives

Abstract

For better classification generative models are used to initialize the model and model features before training a classifier. Typically it is needed to solve separate unsupervised and supervised learning problems. Generative restricted Boltzmann machines and deep belief networks are widely used for unsupervised learning. We developed several supervised models based on DBN in order to improve this two-phase strategy. Modifying the loss function to account for expectation with respect to the underlying generative model, introducing weight bounds, and multi-level programming are applied in model development. The proposed models capture both unsupervised and supervised objectives effectively. The computational study verifies that our models perform better than the two-phase training approach.