Image classification with stacked restricted boltzmann machines and evolutionary function array classification voter

Abstract

Deep belief networks (DBNs) are graphical models which learn to extract the deep hierarchical representation of the input data. Restricted boltzmann machines can be stacked and trained in a greedy manner to form deep belief networks. The combination of deep belief networks and the classifier is used for data classification. However, with the increase of the size of the training set, it requires a huge amount of training time to get better classification accuracy results. To address these issues, a novel fast classifier, called evolutionary function array classification voter (EFACV) is proposed in this research. EFACV is inspired by evolution hardware and it is combined with deep belief networks which is formed by a stack of restricted boltzmann machines for image classification. In our approach, the stacked restricted boltzmann machines are used to extract the feature vectors of the images and EFACV is used to classify these feature vectors. Experimental verifications are conducted on MNIST dataset. Experimental results indicate that compared with the combination of the stacked restricted boltzmann machines and softmax classifier and the combination of the stacked restricted boltzmann machines and support vector machine, our approach can get the higher classification accuracy in less time and have superior anti-overfitting ability.