We'll train the generator by making it better at deceiving the discriminator. To do this, we'll need both networks, similar to the way we trained the discriminator with fake samples:
- We start with a random latent vector, z, and feed it through both the generator and discriminator to produce the output,
. - The loss function is the same as the discriminator loss. However, our goal here is to maximize rather than minimize it, since we want to deceive the discriminator.
- In the backward pass, the discriminator weights,
, are ...
