Abstract: Deep learning has been widely used for network traffic classification, but they are vulnerable to well-designed adversarial examples. From the attacker's point of view, a new method is ...