In this chapter, we took our understanding of the ANNs further, to the deep neural networks that contain more than one, and up to hundreds and thousands of, hidden layers. The learning based on these deep neural networks is called deep learning. Deep learning is evolving as one of the most popular algorithms for solving some of the extremely complex problems within a stochastic environment. We have established the fundamental theory behind the working of deep neural networks and looked at the building blocks of gradient based-learning, backpropagation, nonlinearities, and the regularization technique- dropout. We have also reviewed some of the specialized neural network architecture's CNNs and RNNs.
We have also studied practical approaches for building data preparation pipelines and looked at the examples of applying regularization using the Weka library along with the DataVec library. We have studied some practical approaches for implementing neural network architectures. We have...