Research On Adversarial Attack Algorithms For Network Defense And Attack Applications

Artificial intelligence has outstanding performance in tasks such as image recognition and automatic driving.With the widespread application of artificial intelligence,both attackers and defenders use artificial intelligence tools to achieve their goals more efficiently in the field of network security.However,the adversarial attack algorithm obtains malicious adversarial samples by adding perturbations to the original data,thereby misleading the decision-making of artificial intelligence,which poses a huge challenge to both the attacker and the defender.This paper studies the adversarial attack algorithm from the two perspectives of defense and attack,and shows the problems and challenges faced by both offensive and defensive parties in the field of network security when using artificial intelligence.The main research contents and contributions of this paper are as follows:(1)The attacker uses artificial intelligence to crack the verification code,and this paper proposes an adversarial CAPTCHA generation algorithm from the perspective of defense.First,threat models are established based on the actual CAPTCHA application scenario,and an adversarial algorithm based on random translation and aggregation adversarial perturbation generation is proposed on the basis of the iterative anti-attack algorithm I-FGSM.The adversarial CAPTCHA obtained by adding adversarial perturbations to the original CAPTCHA can effectively defend against cracking by artificial intelligence and the security of CAPTCHA has been improved.(2)The defender uses artificial intelligence to detect abnormal traffic,and this paper proposes an adversarial traffic generation algorithm from the perspective of attack.First,the decision tree is used to sort the traffic features according to the Gini coefficient,and the features that have the greatest impact on the classification task are screened out.Then use the screened features to generate adversarial traffic features,and finally use VAE to generate adversarial traffic,and experiment to verify its functional integrity and attack effectiveness.In two application scenarios in the field of network security,this paper uses the adversarial attack algorithm as a defense method and an attack method respectively to explore its impact on both the attacker and the defender,so as to promote attackers to use artificial intelligence to explore more network security vulnerabilities,and at the same time promote defenders to propose more effective defensive strategies.