Hands-On Artificial Intelligence for Cybersecurity

Discriminator network

We have said that the attack strategy implemented by IDSGAN follows the black-box mode, which means that it is assumed that the attacker has no knowledge of the implementations of the target IDS. In this sense, the discriminator component of IDSGAN tries to mimic the attacked IDS, classifying the output generated by the generator component by comparing it with the normal traffic examples.

In this way, the discriminator is able to provide the necessary feedback to the generator in order to craft the adversarial examples. Therefore, the discriminator component consists of a multilayer neural network whose training dataset contains both the normal traffic and the adversarial examples.

The training phases of the discriminator ...