Adversarial Attacks on SDN-Based Deep Learning IDS System

摘要

In recent years, software defined networking (SDN) has become a novel network architecture and design by employing manageable software between the control and data planes. Many SDN-based intrusion detection systems (IDS) have been proposed in recent researches. On the other hand, deep learning has emerged as an explosive growth in research and industry. The application of deep learning on IDS has been studied in many security-critical scenarios. However, some reports has indicated the vulnerability of adversarial attacks on deep learning IDS system with intentional perturbation injection or manipulation. It is important for the empowered IDS system employing deep learning to be responsible for not leading misclassification. Therefore, exploring the impact on the adversarial attacks over SDN networks or security-critical scenarios applying deep learning is an important step to confront such urgent issues. In this paper, we will conduct the SDN-based experiments on adversarial attacks for deep learning detecting system. We propose a novel class of adversarial attacks that exploits the vulnerability of the deep learning classifiers in SDN environment. Three typical deep learning models combining with four different adversarial testing will be conducted in the simulation for complete analysis.