A gentle introduction to Generative Adversarial Networks, and a practical step-by-step tutorial on making your own with PyTorch.

This beginner-friendly guide will give you hands-on experience:

• learning PyTorch basics
• developing your first PyTorch neural network
• exploring neural network refinements to improve performance
• introduce CUDA GPU acceleration

It will introduce GANs, one of the most exciting areas of machine learning:

• introducing the concept step-by-step, in plain English
• coding the simplest GAN to develop a good workflow
• growing our confidence with an MNIST GAN
• progressing to develop a GAN to generate full-colour human faces
• experiencing how GANs fail, exploring remedies and improving GAN performance and stability

Beyond the very basics, readers can explore more sophisticated GANs:

• convolutional GANs for generated higher quality images
• conditional GANs for generated images of a desired class

The appendices will be useful for students of machine learning as they explain themes often skipped over in many courses:

• calculating ideal loss values for balanced GANs
• probability distributions and sampling them to create images
• carefully chosen examples illustrating how convolutions work
• a brief explanation of why gradient descent isn't suited to adversarial machine learning

All code is available publicly as open source on github.