#Adversarial AI

Adversarial Machine Learning Attacks on CCTV AI: Physical-World Patch Perturbations, Gradient-Based Evasion, and Robust Defenses

A rigorous mathematical deconstruction of physical-world adversarial attacks against CCTV object detectors, covering Projected Gradient Descent (PGD), Expectation Over Transformation (EOT), adversarial camouflage patches, and certified robust defenses.