The discriminator is the easiest part of a GAN structure to understand—the discriminator is going to classify the input image as real or not. This classification will happen in the adversarial training. Essentially, the discriminator will classify the inputs during the forward pass of the neural network. As the generator gets better, it will be harder and harder for the GAN to distinguish between the real and fake images. We monitor the loss functions on the Terminal screen, but we could use them in the future to stop training early.