Deep Learning Methods for Data Science

Abstract

Deep learning network (DLN) is defined as the neural network characterized by complex connected layers to handle a large volume of data, automatic extraction of features, and representation learning for identification and regression problems. This concise chapter on deep learning (DL) methods for data science takes readers through a series of program-writing tasks that introduce them to the use of different DL techniques in various areas of artificial intelligence (AI). It covers zen and tao of the various types of DL methods such as convolutional neural network, recurrent neural network (RNN), denoising autoencoder (DAE), recursive neural network, deep reinforcement learning, deep belief networks (DBNs), and long short-term memory (LSTM), i.e., starting from architecture, learning rules, mathematical model to programing aspects explained in this chapter. The developed and emerging structures of DLN has been applied in applications according to the depth of computational graph, learning, and performance. The knowledge of merits and demerits of each method can train reader toward selection of best suited technique for a given problem statement. For example, the evolution of RNN-based DL architecture innovated many applications in time series, biological, speech-to-text conversion, which has sequence dependent data. RNN handles both real values (time series) and symbolic values of variable length inputs. This chapter covers varieties of application with example to give reader an overall learning. The formulation of this chapter highlights the improvement in applications (such as language, text, signal, and image processing) by modifications in network configuration. This AI technique summarizes the necessity, development, strength, and weakness of DLN models used in data science which will integrate all the basic cores of engineering in near future.