GANs in Action: Deep learning with Generative Adversarial Networks by Jakub Langr and Vladimir Bok. Vladimir is a Data Product Manager at Intent. In addition, I really welcome his statement: Why I Donate All of My Book’s Proceeds to Girls Who Code. Jakub is Co-Found of Hypermile, a UK startup deploying AI across transportation solutions.
Jakub and Vladimir have certainly written a wonderful book on machine learning algorithms that generate realistic imaging. However, this book is really intended for readers who already have some experience with machine learning and neural networks.
Whereas many consumers view new imaging services as a kind of magic, super computing power delivered by AI. There are indeed large machine learning datasets in play that even make this imaging possible.
GANs in Action is a very worthy followup to John Kelleher’s Machine Learning, Melanie Mitchell’s excellent book Artificial Intelligence: A Guide for Thinking Humans, and Sean Gerrish’s How Smart Machines Think. Each author is in fact, addressing in various scales, the introduction to Neural Networks and GANs. Thankfully, Jakub and Vladimir have taken the necessary next step in delivering a wonderful introduction and coding deep dive to GANs.
In fact, for many consumers the Grokking series of books are a must read. Grokking Algorithms: An Illustrated Guide For Programmers and Other Curious People by Aditya Bhargava and Grokking Artificial Intelligence Algorithms: Understand and apply the core algorithms of deep learning and artificial intelligence by Rishal Hurbans. Thus, both are wonderfully illustrated books to begin anyone’s journey into understanding Artificial Intelligence, Machine Learning, and Deep Learning.
Part 1: Introduction to GANs and generative modeling
Jakub and Vladimir certainly deliver an engaging guide addressing Generative Adversarial Network (GANs). In addition they write for the user from the ground up. That said, users should already understand Python objects Pandas and also Keras. Understanding Machine Learning theory will benefit users as well. I certainly need to research backpropagation.They provide insights into new GAN solutions. By seeing how they use tools, users can understand how to develop new applications. Jakub and Vladimir also implement Deep Convolutional GANs (DCGAN).
Part 2: Advanced Topics in GANs
This second part of the book is addressing both theory and practice to ‘train’ GANs. This includes new Progressive GAN (PGGAN) an advanced methodology. PGGANs are now achieving greatly increased image quality. At the same time, they address critical classifications, Semi-Supervised GAN (SGAN), and also introduce Conditional GAN (CGAN). In addition they address Cycle-Consistent Adversarial Networks (CycleGANs). This GAN is very interesting as it has the ability to transform (translate) an image into another image. They use the example of a photograph of a horse transformed into a photograph of a zebra.
Part 3: Where do we go from Here?
What does the future hold for GANs? Jakub and Vladimir outline the future of advances coming to GANs especially as they acknowledge the released of advanced CPUs dedicated to imaging including NVIDIA systems. At the same time, they acknowledge techniques to intentionally deceive Machine Learning models into making mistakes.
I was certainly entertained by the closing chapter addressing medicine (classification accuracy) and fashion (personalization). Finally Jakub and Vladimir address the ethics of GANs. I would highly recommend this as they provide deep insights to GANs and how this Machine Learning is rapidly changing the landscape, for better and for worse. They acknowledge both especially via deep fake videos.
Welcome to your future
In conclusion, this book provides an excellent foundation to GANs and my sincere compliments to their efforts via extensive footnotes and educational links. They also reveal how these tools will lead to new developments in Machine Learning. I have learned so much from this book.