Adversarial Examples Detection for XSS Attacks Based on Generative Adversarial Networks

Xueqin Zhang,Yue Zhou,Jiahao Chen,Jingjing Zhuge,Songwen Pei

doi:10.1109/access.2020.2965184

Abstract

Models based on deep learning are prone to misjudging the results when faced with adversarial examples. In this paper, we propose an MCTS-T algorithm for generating adversarial examples of cross-site scripting (XSS) attacks based on Monte Carlo tree search (MCTS) algorithm. The MCTS algorithm enables the generation model to provide a reward value that reflects the probability of generative examples bypassing the detector. To guarantee the antagonism and feasibility of the generative adversarial examples, the bypassing rules are restricted. The experimental results indicate that the missed detection rate of adversarial examples is significantly improved after the MCTS-T generation algorithm. Additionally, we construct a generative adversarial network (GAN) to optimize the detector and improve the detection rate when dealing with adversarial examples. After several epochs of adversarial training, the accuracy of detecting adversarial examples is significantly improved.

Highlights

Deep learning methods, with their high-precision in classification and high-speed processing performance, are expected to complement or replace traditional intrusion detection technologies in the detection of intrusions under complex internet environments
We proposed a generative adversarial network (GAN)-based adversarial training method named Monte Carlo tree search (MCTS)-T to defend against adversarial examples
This paper proposed an MCTS-T adversarial example generation algorithm for XSS attacks

Summary

INTRODUCTION

With their high-precision in classification and high-speed processing performance, are expected to complement or replace traditional intrusion detection technologies in the detection of intrusions under complex internet environments. Adversarial examples have become an unavoidable problem when developers apply deep learning models to practical issues, especially in the information security field. Adversarial examples take advantage of the highly nonlinear characteristic of the neural network to deceive the model through minor changes to the original samples, making machine learning models produce incorrect classification decisions. X. Zhang et al.: Adversarial Examples Detection for XSS Attacks Based on GANs be done to XSS traffic are restricted in this paper. Using GAN, the discrimination model can avoid traversing all attack samples and accelerate convergence when defending against adversarial examples. We proposed an improved MCTS algorithm to generate adversarial examples of XSS attack traffic data. The proposed MCTS algorithm guarantees the feasibility of the generated adversarial traffic examples and provides a reward for assessing the performance of each bypassing operation. The experimental results showed that our method can effectively obtain adversarial examples, and the GAN-optimized detection model can effectively defend against such adversarial examples

RELATED WORK

EXPERIMENTAL RESULTS AND ANALYSIS

DATASETS AND EVALUATION METRIC

CONCLUSION