Generative Adversarial Nets

Generative Adversarial Networks (GANs) are a framework for estimating generative models via an adversarial process. Two neural networks contest with each other in a game. One network, the generator, produces new data instances, while the other, the discriminator, evaluates them for authenticity. By iteratively training both networks, the generator learns to create increasingly realistic data, while the discriminator becomes better at distinguishing fake from real. This approach provides a way to train generative models without needing Markov chains, using only backpropagation, and can model complex, high-dimensional data.