Hands-On Artificial Intelligence for Cybersecurity

Understanding IDSGAN's algorithm training

To train the generator network, the gradients of the results obtained from the classification of adversarial examples by the discriminator network are used. The objective function, also known as the loss function—represented by, L in the following equation that the generator network must minimize—consists of the following equation:

Here, G and D represent the generator and the discriminator networks, respectively, while S_attack represents the original malicious samples, with M and N representing the m-dimensional vector that matches the original traffic sample and the n-dimensional vector matching ...