Hands-On Artificial Intelligence for Cybersecurity

Generator network

As in all GANs, the generator network is the component responsible for generating the adversarial examples.

In IDSGAN, the generator transforms an original sample of the input traffic, associated with the vector of size m, which represents the characteristics of the original sample, a vector of dimension n, containing noise—that is, random numbers extracted from a uniform distribution whose values fall within the range [0, 1].

The generator network consists of five layers (with which the ReLU activation function is associated) to manage the output of the internal layers, while the output layer has sufficient units to meet the original m-dimensional sample vector.

As we anticipated, the generator network adjusts its parameters ...