Step through one round of GAN training. The discriminator learns to spot fakes, then the generator learns to fool it, one network at a time.
Only one network is updated at a time. The other is held still, so each has a fixed opponent for that step.
