Advanced Deep Learning with Keras Apply deep learning techniques, autoencoders, GANs, variational autoencoders, deep reinforcement learning, policy gradients, and more Rowel Atienza BIRMINGHAM - MUMBAI Advanced Deep Learning with Keras Copyright © 2018 Packt Publishing All rights reserved. No part of this book may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, without the prior written permission of the publisher, except in the case of brief quotations embedded in critical articles or reviews. Every effort has been made in the preparation of this book to ensure the accuracy of the information presented. However, the information contained in this book is sold without warranty, either express or implied. Neither the author, nor Packt Publishing or its dealers and distributors, will be held liable for any damages caused or alleged to have been caused directly or indirectly by this book. Packt Publishing has endeavored to provide trademark information about all of the companies and products mentioned in this book by the appropriate use of capitals. However, Packt Publishing cannot guarantee the accuracy of this information. Acquisition Editor: Frank Pohlmann, Andrew Waldron, Suresh Jain Content Development Editor: Alex Sorrentino Technical Editor: Gaurav Gavas Project Editor: Kishor Rit Proofreader: Safis Editing Indexers: Aishwarya Gangawane Graphics: Tom Scaria Production Coordinator: Sandip Tadge First published: October 2018 Production reference: 1311018 Published by Packt Publishing Ltd. Livery Place 35 Livery Street Birmingham B3 2PB, UK. ISBN 978-1-78862-941-6 www.packtpub.com mapt.io Mapt is an online digital library that gives you full access to over 5,000 books and videos, as well as industry leading tools to help you plan your personal development and advance your career. For more information, please visit our website. Why subscribe? • Spend less time learning and more time coding with practical eBooks and Videos from over 4,000 industry professionals • Learn better with Skill Plans built especially for you • Get a free eBook or video every month • Mapt is fully searchable • Copy and paste, print, and bookmark content Packt.com Did you know that Packt offers eBook versions of every book published, with PDF and ePub files available? You can upgrade to the eBook version at www.Packt. com and as a print book customer, you are entitled to a discount on the eBook copy. Get in touch with us at [email protected] for more details. At www.Packt.com, you can also read a collection of free technical articles, sign up for a range of free newsletters, and receive exclusive discounts and offers on Packt books and eBooks. Contributors About the author Rowel Atienza is an Associate Professor at the Electrical and Electronics Engineering Institute of the University of the Philippines, Diliman. He holds the Dado and Maria Banatao Institute Professorial Chair in Artificial Intelligence. Rowel has been fascinated with intelligent robots since he graduated from the University of the Philippines. He received his MEng from the National University of Singapore for his work on an AI-enhanced four-legged robot. He finished his Ph.D. at The Australian National University for his contribution on the field of active gaze tracking for human-robot interaction. Rowel's current research work focuses on AI and computer vision. He dreams on building useful machines that can perceive, understand, and reason. To help make his dreams become real, Rowel has been supported by grants from the Department of Science and Technology (DOST), Samsung Research Philippines, and Commission on Higher Education-Philippine California Advanced Research Institutes (CHED-PCARI). I would like to thank my family, Che, Diwa, and Jacob. They never cease to support my work. I would like to thank my mother who instilled into me the value of education. I would like to express my gratitude to the people of Packt and this book's technical reviewer, Frank, Kishor, Alex, and Valerio. They are inspiring and easy to work with. I would like to thank the institutions who always support my teaching and research agenda, University of the Philippines, DOST, Samsung Research PH, and CHED-PCARI. I would like to acknowledge my students. They have been patient as I develop my courses in AI. About the reviewer Valerio Maggio is currently a Post-Doc Data Scientist at Fondazione Bruno Kessler (FBK) in Trento, Italy, responsible for Machine Learning and Deep Learning in the MPBA lab (Predictive Models for Biomedicine and Environment). Valerio has a Ph.D. in Computational Science from the University of Naples "Federico II." His research interests are focused on Machine Learning and Deep Learning applied to Software Maintenance and Computational Biology. Valerio is very much involved in the scientific Python community, and he is an active speaker at many Python conference. He is also the lead organiser of PyCon Italy/PyData Florence, and EuroSciPy. He uses Python as the mainstream language for his deep/machine learning code, making an intensive use of Python to analyse, visualise, and learn from data. In the context of Deep Learning, Valerio is the author of a quite popular Keras/TensorFlow tutorial, publicly available on his GitHub Profile – github.com/leriomaggio/deep- learning-keras-tensorflow – and presented in many conferences (EuroSciPy, PyData London, PySS) and University courses. Valerio is also passionate about (black) tea, and an "old-school" Magic The Gathering (MTG) player, who enjoys playing and teaching MTG to newbies. Packt is Searching for Authors Like You If you're interested in becoming an author for Packt, please visit authors.packtpub. com and apply today. We have worked with thousands of developers and tech professionals, just like you, to help them share their insight with the global tech community. You can make a general application, apply for a specific hot topic that we are recruiting an author for, or submit your own idea. Table of Contents Preface V Chapter 1: Introducing Advanced Deep Learning with Keras 1 Why is Keras the perfect deep learning library? 2 Installing Keras and TensorFlow 3 Implementing the core deep learning models - MLPs, CNNs and RNNs 4 The difference between MLPs, CNNs, and RNNs 5 Multilayer perceptrons (MLPs) 6 MNIST dataset 6 MNIST digits classifier model 8 Building a model using MLPs and Keras 12 Regularization 14 Output activation and loss function 15 Optimization 17 Performance evaluation 20 Model summary 21 Convolutional neural networks (CNNs) 23 Convolution 26 Pooling operations 27 Performance evaluation and model summary 28 Recurrent neural networks (RNNs) 31 Conclusion 37 Chapter 2: Deep Neural Networks 39 Functional API 40 Creating a two-input and one-output model 43 [ i ] Table of Contents Deep residual networks (ResNet) 49 ResNet v2 59 Densely connected convolutional networks (DenseNet) 62 Building a 100-layer DenseNet-BC for CIFAR10 65 Conclusion 68 References 69 Chapter 3: Autoencoders 71 Principles of autoencoders 72 Building autoencoders using Keras 74 Denoising autoencoder (DAE) 84 Automatic colorization autoencoder 89 Conclusion 97 References 97 Chapter 4: Generative Adversarial Networks (GANs) 99 An overview of GANs 99 Principles of GANs 100 GAN implementation in Keras 105 Conditional GAN 114 Conclusion 124 References 124 Chapter 5: Improved GANs 125 Wasserstein GAN 126 Distance functions 126 Distance function in GANs 128 Use of Wasserstein loss 131 WGAN implementation using Keras 135 Least-squares GAN (LSGAN) 142 Auxiliary classifier GAN (ACGAN) 147 Conclusion 160 References 160 Chapter 6: Disentangled Representation GANs 161 Disentangled representations 162 InfoGAN 164 Implementation of InfoGAN in Keras 167 Generator outputs of InfoGAN 177 StackedGAN 179 Implementation of StackedGAN in Keras 181 Generator outputs of StackedGAN 197 Conclusion 200 Reference 201 [ ii ] Table of Contents Chapter 7: Cross-Domain GANs 203 Principles of CycleGAN 204 The CycleGAN Model 207 Implementing CycleGAN using Keras 211 Generator outputs of CycleGAN 225 CycleGAN on MNIST and SVHN datasets 227 Conclusion 234 References 235 Chapter 8: Variational Autoencoders (VAEs) 237 Principles of VAEs 238 Variational inference 239 Core equation 240 Optimization 241 Reparameterization trick 242 Decoder testing 243 VAEs in Keras 244 Using CNNs for VAEs 249 Conditional VAE (CVAE) 254 β-VAE: VAE with disentangled latent representations 264 Conclusion 268 References 268 Chapter 9: Deep Reinforcement Learning 271 Principles of reinforcement learning (RL) 272 The Q value 274 Q-Learning example 276 Q-Learning in Python 281 Nondeterministic environment 287 Temporal-difference learning 287 Q-Learning on OpenAI gym 288 Deep Q-Network (DQN) 293 DQN on Keras 296 Double Q-Learning (DDQN) 302 Conclusion 304 References 305 Chapter 10: Policy Gradient Methods 307 Policy gradient theorem 308 Monte Carlo policy gradient (REINFORCE) method 311 REINFORCE with baseline method 313 Actor-Critic method 315 Advantage Actor-Critic (A2C) method 317 [ iii ]