Generative adversarial networks and adversarial training are truly limitless in concept but often fall short in execution and implementation. As we have seen throughout this book, the failures often reside in the generator. And, as we have learned, the key to a good GAN is a good generator.
The discriminator in a GAN is simply a classifier and often a simple binary one at that. In fact, it is a common practice now to build a GAN to train a discriminator/classifier. A GAN will be trained on a set of data for binary classification, and after training, the generator is discarded. Not only ...
