Unsupervised Deep Learning Architectures

Abstract

The cascade of multiple layers of a deep learning architecture can be learnt in an unsupervised manner for the tasks like pattern analysis. A deep learning architecture can be trained one layer at a time, treating each layer in turn as an unsupervised restricted Boltzmann machine. Unsupervised deep learning algorithms are important because unlabeled data is more abundant than the labeled data. For applications with large volumes of unlabeled data, a two-step procedure is used: in the first step, a deep neural network is pretrained in an unsupervised manner; in the second step, a small portion of the unlabeled data is manually labeled, and then used for supervised fine-tuning of the deep neural network.