Organization of the Book
This book consists of six chapters, which can be grouped into three subjects. The first subject is Machine Learning and takes place in Chapter 1. Deep Learning stems from Machine Learning. This implies that if you want to understand the essence of Deep Learning, you have to know the philosophy behind Machine Learning to some extent. Chapter 1 starts with the relationship between Machine Learning and Deep Learning, followed by problem solving strategies and fundamental limitations of Machine Learning. The detailed techniques are not introduced in this chapter. Instead, fundamental concepts that apply to both the neural network and Deep Learning will be covered.
The second subject is the artificial neural network.1 Chapters 2-4 focus on this subject. As Deep Learning is a type of Machine Learning that employs a neural network, the neural network is inseparable from Deep Learning. Chapter 2 starts with the fundamentals of the neural network: principles of its operation, architecture, and learning rules. It also provides the reason that the simple single-layer architecture evolved to the complex multi-layer architecture. Chapter 3 presents the back-propagation algorithm, which is an important and representative learning rule of the neural network and also employed in Deep Learning. This chapter explains how cost functions and learning rules are related and which cost functions are widely employed in Deep Learning.
Chapter 4 explains how to apply the neural network to classification problems. We have allocated a separate section for classification because it is currently the most prevailing application of Machine Learning. For example, image recognition, one of the primary applications of Deep Learning, is a classification problem.
The third topic is Deep Learning. It is the main topic of this book. Deep Learning is covered in Chapters 5 and 6. Chapter 5 introduces the drivers that enable Deep Learning to yield excellent performance. For a better understanding, it starts with the history of barriers and solutions of Deep Learning. Chapter 6 covers the convolution neural network, which is representative of Deep Learning techniques. The convolution neural network is second to none in terms of image recognition. This chapter starts with an introduction of the basic concept and architecture of the convolution neural network as it compares with the previous image recognition algorithms. It is followed by an explanation of the roles and operations of the convolution layer and pooling layer, which act as essential components of the convolution neural network. The chapter concludes with an example of digit image recognition using the convolution neural network and investigates the evolution of the image throughout the layers.