Ch:14 General Adversarial Networks (GAN's) with Math.

Mode collapse happens quite often and there are some ways to prevent it from happening, #willdiscusshortly. This is a very often problem we see in deep neural networks in general, the same problem gets stronger here because the gradient at Discriminator not only goes back to Discriminator network but also it goes back to Generator network as feedback. Because of it there is no stability in training GAN's. Nash equilibrium happens when one player does not change his/her actions regardless of what the other is doing. Here the kid is neither lossing nor winning.