Big-Data Analytics for Cloud, IoT and Cognitive Computing

6Deep Learning with Artificial Neural Networks

CHAPTER OUTLINE

1. 6.1 Introduction
   1. 6.1.1 Deep Learning Mimics Human Senses
   2. 6.1.2 Biological Neurons versus Artificial Neurons
   3. 6.1.3 Deep Learning versus Shallow Learning
2. 6.2 Artificial Neural Networks (ANN)
   1. 6.2.1 Single Layer Artificial Neural Networks
   2. 6.2.2 Multilayer Artificial Neural Network
   3. 6.2.3 Forward Propagation and Back Propagation in ANN
3. 6.3 Stacked AutoEncoder and Deep Belief Network
   1. 6.3.1 AutoEncoder
   2. 6.3.2 Stacked AutoEncoder
   3. 6.3.3 Restricted Boltzmann Machine
   4. 6.3.4 Deep Belief Networks
4. 6.4 Convolutional Neural Networks (CNN) and Extensions
   1. 6.4.1 Convolution in CNN
   2. 6.4.2 Pooling in CNN
   3. 6.4.3 Deep Convolutional Neural Networks
   4. 6.4.4 Other Deep Learning Networks
5. 6.5 Conclusions

6.1 Introduction

Deep learning simulates the operations in layers of artificial neural networks. This is heavily used to extract and learn features from data. A multilayer neural network includes one input layer, one output layer and multiple hidden layers. The connection strength between neurons is adjusted in the learning process. Common deep learning architectures are introduced in subsequent sections, including the basic ANN, convolutional neural networks, deep belief network and recurrent neural network, etc.

6.1.1 Deep Learning Mimics Human Senses

In March 2016, AlphaGo generated high publicity and debate on human-machine competition in intelligence. After five rounds of competition, finally the computer beat Sedol Lee, ...