GANs were introduced by a group of researchers at the University of Montreal led by Ian Goodfellow. The core idea behind a GAN model is to have two competing neural network models. One network takes the noise as input and generates samples (hence known as generator). The second model (known as discriminator) gets samples from both the generator and the actual training data, and should be able to differentiate between the two sources. Generative and discriminative networks are playing a continuous game, where the generator model is learning to generate more realistic samples or examples, and the discriminator is learning to get better and better at differentiating generated data from the real data. The two networks are trained simultaneously, and the goal is that the competition will make the generated samples indistinguishable from the real data:
The analogy used to describe GANs is that the generator is like a forger that is attempting to produce some forged material, and the discriminator...