EMShepherd: Detecting Adversarial Samples via Side-channel Leakage

Ruyi Ding, Cheng Gongye, Siyue Wang +2

Deep Neural Networks (DNN) are vulnerable to adversarial perturbations-small changes crafted deliberately on the input to mislead the model for wrong predictions. Adversarial attac…