面向网络入侵检测的对抗样本攻击与防御技术综述OA

With its powerful reasoning and recognition ability,machine learning technology and deep learning technology have promoted the rapid development of network intrusion detection system(NIDS)based on anomaly detection.However,these technologies have inherent defects and are more vulnerable to adversarial example(AE)attacks.AE will add some small disturbances to the original data,which are difficult for human eyes to detect,making NIDS misjudged.Aiming at the AE problem in NIDS,this study systematically sorts out and analyzes the latest research progress from the two aspects of attack methods and protection strategies.Firstly,the operation mechanism of NIDS and the generation mechanism of AE are explained,and the theoretical framework of offensive and defensive confrontation is constructed.Then,a multi-dimensional classification system is constructed based on the core characteristics of AE,such as concealment and migra-tion.The existing research results are compared and sorted out from the perspective of attack and defense.Finally,the evo-lution of technology and the current situation of practical application are sorted out,and the future development direction is discussed.