Abstract: Adversarial attacks have exposed serious vulnerabilities in deep neural networks (DNNs), causing misclassifications through human-imperceptible perturbations to DNN inputs. We explore a new ...