Abstract: Studies have shown that deep learning models are vulnerable to adversarial examples, which cause incorrect predictions by adding imperceptible perturbations into normal inputs. The same ...