Artificial Intelligence Programming with Python

CHAPTER 9GAN and Neural-Style Transfer

“A year spent in artificial intelligence is enough to make one believe in God.”

—Alan Perlis (American computer scientist)

CHAPTER OUTLINE

• 9.1 Introduction
• 9.2 Generative Adversarial Networks
• 9.3 Neural-Style Transfer
• 9.4 Adversarial Machine Learning
• 9.5 Music Generation
• 9.6 Summary
• 9.7 Chapter Review Questions

9.1 Introduction

Over the years, there have been many exciting developments in the field of deep learning, such as generative adversarial networks (GANs) and neural-style transfer. In this chapter, we will first introduce GAN and neural-style transfer and then introduce adversarial machine learning and music generation.

9.2 Generative Adversarial Network

A generative adversarial network (GAN) is probably one of the most exciting developments in deep learning. A GAN is a class of new deep learning neural networks designed by Ian Goodfellow and his colleagues from the Université de Montréal in 2014. Ian Goodfellow was doing his PhD in machine learning under the supervision of Yoshua Bengio and Aaron Courville.

In a GAN architecture, two neural networks (agents) compete against each other in the form of a zero-sum game, where one agent's gain is another agent's loss. A GAN is an approach to generate new images using deep learning methods, such as convolutional neural networks.

Figure 9.1 shows the schematic diagram of a GAN, including two key components: Generator and Discriminator. In this example, both Generator and ...