Deep Convolutional GANs

In the previous sections, where we implemented a GAN, we made use of an architecture based on the Multi-Layer Perceptron (MLP). As you may recall from the previous chapters, MLPs have fully connected layers. This implies that all the neurons in each layer have connections to all the neurons of the subsequent layer. For this reason, MLPs are also called fully connected layers. The GAN that we developed in the previous section can also be called a Fully Connected GAN (FCGAN). In this section, we will learn about another architecture called Deep Convolutional GANs (DCGANS). As the name implies, this is based on the Convolutional Neural Network (CNN) architecture that you learned about in Chapter 4, Deep Learning for Text – Embeddings. Let's revisit some of the building blocks of DCGANs.

Building Blocks of DCGANs

Most of the building blocks of DCGANs are similar to what you learned about when you were introduced to CNNs in Chapter 3, Image Classification...