Generative Adversarial Networks .

Abstract

We propose a new framework for estimating generative models via an adversarial process: a generator that creates fake samples, and a discriminator that tries to distinguish real from fake.

Game Theory Framework

• Generator G: Tries to fool the discriminator, minimizes log(1 - D(G(z)))
• Discriminator D: Tries to correctly classify real vs fake, maximizes log(D(x))

They play a minimax game until Nash equilibrium.

Impact

GANs enabled photorealistic image generation, style transfer, and countless creative applications. Yann LeCun called them "the most interesting idea in deep learning in the last 10 years."