Design of GAN

 

Designing the network architectures #

The architecture of the generator network is described in the following diagram. It takes a 1-dimension random value as input and gives a 10-dimension vector as output. It has 2 hidden layers, with each containing 10 neurons. The calculation in each layer is a matrix multiplication. Therefore, the network is, in fact, a multilayer perceptron (MLP):

The architecture of the discriminator network is described in the following diagram. It takes a 10-dimension vector as input and gives a 1-dimension value as output. The output is the prediction label (real or fake) of the input sample. The discriminator network is also an MLP with two hidden layers, each containing 10 neurons: