ABSTRACT
In this chapter, we have discussed three major areas of design of intrusion detection systems (IDS) where generative adversarial networks (GANs) are used. First, we have discussed the current research work, which addresses the problem of imbalanced dataset using GANs wherein the minority class samples are oversampled using the generator of a GAN to mitigate the imbalance. Second, we have discussed current research work, which focuses on the use of GANs as an anomaly detector by using the reconstruction and discrimination losses for computing anomaly score. The anomaly detection framework of GAN learns the distribution of the benign class data samples and consequently yields a higher reconstruction and discrimination loss if it is fed with a malign class data sample. Third, we have discussed some established results in the field of adversarial machine learning, where GANs are trained to generate adversarial examples against machine learning/deep learning-based IDS. This type of research aims at devising methods for verifying the robustness of a machine learning-based IDS.
