Hands-On Artificial Intelligence for Cybersecurity

How to train a GAN

Training a GAN may require high computational capacity; otherwise, the time required to carry out the training may vary from a few hours to a few days. Given the mutual dependency that is established between the two NNs, it is advisable to keep the values returned by the generator network constant while training the discriminator network. At the same time, it can be useful to perform the pretraining of the discriminator network using the training data available, before training the generator network.

It is also important to adequately set the learning rates of the two NNs, so as to avoid a situation where the learning rate of the discriminator network exceeds that of the generator network and vice versa, thereby preventing ...