Abstract: The well-trained image classification neural networks are vulnerable to adversarial examples. An adversarial example is a malicious input carefully crafted by adding small perturbations to ...