Generative Adversarial Networks. A relatively new innovation in the field of deep learning is the Generative Adversarial Network (GAN). GANs use two different networks, one that generates images of what it thinks are shoes (this is called the generator), and another that tries to distinguish between the fake shoes and real shoes from the data set (this is called the discriminator). Think of it as a Cold War between these networks, except they’re trying to outsmart each other, not out-arm each other. Once the discriminator basically can’t tell the difference between the generated images and the real images, the model has converged.