XAI to Explore Robustness of Features in Adversarial Training for Cybersecurity

Adversarial training is an effective learning approach to harden deep neural models against adversarial examples. In this paper, we explore the accuracy of adversarial training in cybersecurity. In addition, we use an XAI technique to analyze how certain input features may have an effect on decisions yielded with adversarial training giving the security analyst much better insight into robustness of features. Finally, we start the investigation of how XAI can be used for robust features selection within adversarial training in cybersecurity problems.