Post-train Black-box Defense via Bayesian Boundary Correction

He Wang, Yunfeng Diao

Classifiers based on deep neural networks are susceptible to adversarial attack, where the widely existing vulnerability has invoked the research in defending them from potential t…