Abstract: Numerous researches on adversarial black-box attacks have proved that deep neural networks have certain insecurity. However, the current black-box attack methods still have shortages in ...