An adversarial sample generation model for the intrusion detection system based on Generative Adversarial Networks

DOI:	10.11809/bqzbgcxb2023.06.027
Keywords:	intrusion detection system; generative adversarial networks; adversarial samples; deep learning; black box attack; feature constraint
Abstract:	With the rise of artificial intelligence technology, the network intrusion detection system based on deep learning has been widely used, but neural network models are easily affected by adversarial disturbances. The attackers build adversarial samples by adding small perturbations to the network traffic so that the intrusion detection system can misclassify them. Based on the design and improvement of Generative Adversarial Networks (GAN), this paper proposes an adversarial sample generation model called AdvWGAN, which can generate adversarial malicious traffic that meets the characteristics of the network traffic, and carry out adversarial attacks on the black box intrusion detection system. The experiments show that AdvWGAN can implement effective black box attacks on the intrusion detection models based on deep learning on the premise of ensuring the real and effective network traffic.
Issue:	Vol. 44 No. 6 (2023)
Published:	2023-06-28
PDF HTML