Abstract: Deep neural network (DNN) has strong representation learning ability, but it is vulnerable and easy to be fooled by adversarial examples. In order to handle the vulnerability of DNN, many ...