A comprehensive study of deep neural networks for unsupervised deep learning

Abstract

Deep learning methods aims at learning meaningful representations in the field of machine learning (ML). Unsupervised deep learning architectures has grown at a fast pace owing to their ability to learn intricate problems. Availability of large amount of labelled and unlabeled data with highly efficient computational resources makes deep learning models more practicable for different applications. Recently, deep neural networks (DNNs) have become an extremely effective and widespread research area in the field of machine learning. The significant aim of deep learning is to learn the primary structure of input data and also to investigate the nonlinear mapping between the inputs and outputs. The main emphasis of this chapter is on unsupervised deep learning. We first study difficulties with neural networks while training with backpropagation-algorithms. Later, different structures, namely, restricted Boltzmann machines (RBMs), Deep Belief Networks (DBNs), nonlinear autoencoders, deep Boltzmann machines are covered. Lastly, sustainable real applications in agricultural domain with deep learning are described.