Author: Jakub Langr and Vladimir Bok
Publisher: Manning
Date: October 2019
Pages: 240
ISBN: 978-1617295560
Print: 1617295566
Audience: Data professionals with intermediate Python skills
Rating: 4
Reviewer: Mike James
GANs are both fun and potentially profitable. Can you get started with such a slim book?

GANs - Generative Adversarial Networks have been used to do all sorts of interesting things from creating faces of people who don't exist to creating paintings in the style of any famous painter.  We have reported many examples see Christies To Auction AI-Generated Artwork, GauGAN Will Draw Your Landscape For You and Using GANs For Underwater Color Images.

Even if you are not an AI expert you might well want to get to grips with how they work and how to implement one and this book might be one way of doing just that. It isn't deeply mathematical, but it doesn't avoid all of the math. It isn't deeply hands on, but it does tell you how to implement different types of GAN using Python, Juypter Notebooks and Keras. It is also very informal and mostly easy to read. However, if you are looking for a book that defines everything in detail and is mathematically rigorous this will not be the book for you. Sometimes the omission of technical and mathematical detail means that you have to look elsewhere to make the idea precise, but this is the cost you must pay for a short and informal introduction.

After a brief introduction to the idea of a GAN we move on to the basics of the autoencoder - a neural network that learns to reproduce its inputs while using only a small number of neurons to represent the data. The first practical is to implement an autoencoder. In the next chapter we implement a GAN to generate numerals based on tje MNIST data. By this point in the book you will discover that you are expected to know a fair amount about neural networks and their architecture. This is not a first book for the complete neural network beginner. The final chapter of the section is about Deep Convolutional GANs

Part 2 is called "Advanced Topics in GANs" and it does go into a great many details that are normally ignored. First we have a chapter on training and things that can go wrong. Next a chapter on various techniques for improving GANs including using TensorFlow Hub. The final three chapters are on semi-supervised GANs, conditional GANs and CycleGAN.

The final part of the book is a more geneal look at what you can use GANs for in medicine and fashion and a look ahead at what remeains to be done.

As already mentioned, this is a short book and doesn't tell you all the details, but if you have the time to read it and the resources to follow the practicals, you should emerge as a GAN expert. If you are looking for an in-depth technical introduction then this isn't it. It also isn't suitable for the reader who has no background in neural networks. If you fit the profile you should enjoy reading this book.

To keep up with our coverage of books for programmers, follow @bookwatchiprog on Twitter or subscribe to I Programmer's Books RSS feed for each day's new addition to Book Watch and for new reviews.