Generative Adversarial Networks (GANs)
A beginner’s guide on GANs can be found here.
Unlike traditional machine learning where we predict labels given a set of features, in GANs, we predict features given the labels. This is also the main difference between generative and discriminative algorithms.
How do GANs work?
Two networks are beating each other:
- A discriminator tries to identify the instances coming from the generator as fake.
- A generator trying to generate passable instances, to lie without being caught.
Image credit: Thalles Silva
Generative algorithms:
- Given a label, they predict the associated features (Naive Bayes)
- Given a hidden representation, they predict the associated features (VAE, GAN)
- Given some of the features, they predict the rest (inpainting, imputation)