Introduction

Yann LeCun described it as “the most interesting idea in the last 10 years in Machine Learning”. Of course, such a compliment coming from such a prominent researcher in the deep learning area is always a great advertisement for the subject we are talking about! And, indeed, Generative Adversarial Networks (GANs for short) have had a huge success since they were introduced in 2014 by Ian J. Goodfellow and co-authors in the article Generative Adversarial Nets.

So what are Generative Adversarial Networks ? What makes them so “interesting” ? In this post, we will see that adversarial training is an enlightening idea, beautiful by its simplicity, that represents a real conceptual progress for Machine Learning and more especially for generative models (in the same way as backpropagation is a simple but really smart trick that made the ground idea of neural networks became so popular and efficient).

Before going into the details, let’s give a quick overview of what GANs are made for. Generative Adversarial Networks belong to the set of generative models. It means that they are able to produce / to generate (we’ll see how) new content. To illustrate this notion of “generative models”, we can take a look at some well known examples of results obtained with GANs.

Naturally, this ability to generate new content makes GANs look a little bit “magic”, at least at first sight. In the following parts, we will overcome the apparent magic of GANs in order to dive into ideas, maths and modelling behind these models. Not only we will discuss the fundamental notions Generative Adversarial Networks rely on but, more, we will build step by step and starting from the very beginning the reasoning that leads to these ideas.

‍