A Distillation Learning Model of Adaptive Structural Deep Belief Network for AffectNet: Facial Expression Image Database

TL;DR

This paper introduces an adaptive structural deep belief network with a distillation learning approach to improve facial expression classification on AffectNet, achieving significant accuracy gains over traditional models.

Contribution

The paper presents a novel adaptive DBN with neuron generation-annihilation and a distillation learning method to handle ambiguous facial expressions.

Findings

Classification accuracy improved from 78.4% to 91.3%.

Adaptive structure learning effectively discovers optimal network architecture.

Distillation learning enhances model performance on ambiguous cases.

Abstract

Deep Learning has a hierarchical network architecture to represent the complicated feature of input patterns. We have developed the adaptive structure learning method of Deep Belief Network (DBN) that can discover an optimal number of hidden neurons for given input data in a Restricted Boltzmann Machine (RBM) by neuron generation-annihilation algorithm, and can obtain the appropriate number of hidden layers in DBN. In this paper, our model is applied to a facial expression image data set, AffectNet. The system has higher classification capability than the traditional CNN. However, our model was not able to classify some test cases correctly because human emotions contain many ambiguous features or patterns leading wrong answer by two or more annotators who have different subjective judgment for a facial image. In order to represent such cases, this paper investigated a distillation learning model of Adaptive DBN. The original trained model can be seen as a parent model and some child models are trained for some mis-classified cases. For the difference between the parent model and the child one, KL divergence is monitored and then some appropriate new neurons at the parent model are generated according to KL divergence to improve classification accuracy. In this paper, the classification accuracy was improved from 78.4% to 91.3% by the proposed method.