GAN - Generative Adversarial Network

In short: training a model by using two networks; each network creates a worst-case-scenario input for the other network.

What’s an adversarial network?

Firstly, an adversarial network: a network with two players, each player is a neural net. The two networks in some way “compete” against eachother, by producing a worst-case-scenario input for the other network. The network(s) then become optimized for handling these worst-case-scenarios.

GANs are hard to train.