Abstract: Deep neural networks yield desirable performance in text, image, and speech classification. However, these networks are vulnerable to adversarial examples. An adversarial example is a sample ...