Description
Adversarial Artificial Intelligence (AAI). Evasion attacks. Poisoning attacks. Data injection attacks. Data modification attacks. Label flipping. White-box attacks (FGSM – Fast Gradient Sign Method, PGD – Projected Gradient Descent, ACG – Auto-Conjugate Gradient, C&W – Carlini & Wagner, BB – Brendel and Bethge). Adversarial examples in security applications. Defense paradigms against AAI. Adversarial training. Data compression. Model modification. Regularization (Parseval networks). Gradient masking. Defensive distillation. Supporting tools (CleverHans, Adversarial Robustness Toolbox). Practical applications of AAI. Traditional computer networks. Cyber-physical systems in critical infrastructures.