Generative Adversarial Networks

TL;DR

This paper introduces Generative Adversarial Networks (GANs), a novel framework where two neural networks compete to generate and discriminate data, enabling realistic data synthesis without complex inference procedures.

Contribution

The paper presents the GAN framework, a new adversarial training method for generative models that is simple, effective, and does not require Markov chains or approximate inference.

Findings

GANs can generate realistic data samples.

The training process is a minimax game between generator and discriminator.

The framework is theoretically sound with a unique equilibrium solution.

Abstract

We propose a new framework for estimating generative models via an adversarial process, in which we simultaneously train two models: a generative model G that captures the data distribution, and a discriminative model D that estimates the probability that a sample came from the training data rather than G. The training procedure for G is to maximize the probability of D making a mistake. This framework corresponds to a minimax two-player game. In the space of arbitrary functions G and D, a unique solution exists, with G recovering the training data distribution and D equal to 1/2 everywhere. In the case where G and D are defined by multilayer perceptrons, the entire system can be trained with backpropagation. There is no need for any Markov chains or unrolled approximate inference networks during either training or generation of samples. Experiments demonstrate the potential of the framework through qualitative and quantitative evaluation of the generated samples.