Generative Adversarial Networks
This work proposes a foundational new paradigm for generative modeling in machine learning, enabling efficient sample generation across various domains.
The authors tackled the problem of estimating generative models by introducing a framework that trains a generative model and a discriminative model adversarially, resulting in a system that can recover the training data distribution without needing Markov chains or approximate inference networks.
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.