Book Image

Advanced Deep Learning with Keras

By : Rowel Atienza
Book Image

Advanced Deep Learning with Keras

By: Rowel Atienza

Overview of this book

Recent developments in deep learning, including Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), and Deep Reinforcement Learning (DRL) are creating impressive AI results in our news headlines - such as AlphaGo Zero beating world chess champions, and generative AI that can create art paintings that sell for over $400k because they are so human-like. Advanced Deep Learning with Keras is a comprehensive guide to the advanced deep learning techniques available today, so you can create your own cutting-edge AI. Using Keras as an open-source deep learning library, you'll find hands-on projects throughout that show you how to create more effective AI with the latest techniques. The journey begins with an overview of MLPs, CNNs, and RNNs, which are the building blocks for the more advanced techniques in the book. You’ll learn how to implement deep learning models with Keras and TensorFlow 1.x, and move forwards to advanced techniques, as you explore deep neural network architectures, including ResNet and DenseNet, and how to create autoencoders. You then learn all about GANs, and how they can open new levels of AI performance. Next, you’ll get up to speed with how VAEs are implemented, and you’ll see how GANs and VAEs have the generative power to synthesize data that can be extremely convincing to humans - a major stride forward for modern AI. To complete this set of advanced techniques, you'll learn how to implement DRL such as Deep Q-Learning and Policy Gradient Methods, which are critical to many modern results in AI.
Table of Contents (16 chapters)
Advanced Deep Learning with Keras
Contributors
Preface
Other Books You May Enjoy
Index

Least-squares GAN (LSGAN)


As discussed in the previous section, the original GAN is difficult to train. The problem arises when the GAN optimizes its loss function; it's actually optimizing the Jensen-Shannon divergence, DJS. It is difficult to optimize DJS when there is little to no overlap between two distribution functions.

WGAN proposed to address the problem by using the EMD or Wasserstein 1 loss function which has a smooth differentiable function even when there is little or no overlap between the two distributions. However, WGAN is not concerned with the generated image quality. Apart from stability issues, there are still areas of improvement in terms of perceptive quality in the generated images of the original GAN. LSGAN theorizes that the twin problems can be solved simultaneously.

LSGAN proposes the least squares loss. Figure 5.2.1 demonstrates why the use of a sigmoid cross entropy loss in the GAN results in poorly generated data quality. Ideally, the fake samples distribution...