As promised, we will be taking a closer look at variants of the GANs we mentioned in detail in the previous sections and apply them on real-world problems. The most commonly used GANs include deep convolutional GANs, conditional GANs, and the information-maximizing GANs. Let's start with the most basic architecture.
GAN architectures and implementations
Vanilla GANs
In a most basic GAN model, both the generator and discriminator are fully connected neural networks. The architecture of a vanilla GAN can be depicted as follows:
The input of the generator is the random samples from a particular distribution, which we usually call noise or latent variables. The second layer and several ones after are the hidden...
