Threats to Adversarial Training for IDSs and Mitigation - Archive ouverte HAL Access content directly
Conference Papers Year : 2022

Threats to Adversarial Training for IDSs and Mitigation

Abstract

Intrusion Detection Systems (IDS) are essential tools to protect network security from malicious traffic. IDS have recently made significant advancements in their detection capabilities through deep learning algorithms compared to conventional approaches. However, these algorithms are susceptible to new types of adversarial evasion attacks. Deep learning-based IDS, in particular, are vulnerable to adversarial attacks based on Generative Adversarial Networks (GAN). First, this paper identifies the main threats to the robustness of IDS against adversarial sample attacks that aim at evading IDS detection by focusing on potential weaknesses in the structure and content of the dataset rather than on its representativeness. In addition, we propose an approach to improve the performance of adversarial training by driving it to focus on the best evasion candidates samples in the dataset. We find that GAN adversarial attack evasion capabilities are significantly reduced when our method is used to strengthen the IDS.

Dates and versions

hal-03801012 , version 1 (06-10-2022)

Identifiers

Cite

Hassan Chaitou, Jean Leneutre, Laurent Pautet. Threats to Adversarial Training for IDSs and Mitigation. 19th International Conference on Security and Cryptography, Jul 2022, Lisbon, France. pp.226-236, ⟨10.5220/0011277600003283⟩. ⟨hal-03801012⟩
15 View
0 Download

Altmetric

Share

Gmail Facebook Twitter LinkedIn More