SITRAN: Self-Supervised IDS With Transferable Techniques for 5G Industrial Environments
Hyunjin Kim, Jonghoon Lee, Jong‐Geun Park · IEEE Internet of Things Journal · 2024
The evolution of 5G mobile communication technology, in use for over four years, has rapidly increased the global subscriber base. This technological progress extends beyond broadcasting and mobile communications to various industries, particularly through the integration of 5G into smart factories. This integration has shifted the traditional production paradigm from mass production to customized production systems that meet individual customer demands. However, the digitization of smart factories has expanded their attack surfaces, making them potential targets for cyber threats. Despite the development of various security solutions, threats are becoming increasingly sophisticated, necessitating research into AI-based threat detection technologies. AI-based threat detection in real-world scenarios faces challenges in collecting sufficient datasets and deploying detection models in resource-constrained environments. Therefore, we propose a self-supervised learning-based network intrusion detection system (NIDS) that reduces reliance on labeled data and utilizes lightweight models for deployment through knowledge distillation techniques. Our system achieves continuous learning by transferring knowledge between teacher and student models. To validate its efficiency and applicability, experiments were conducted using simulated hacking datasets collected from a 5G smart factory testbed and evaluated using standard benchmark datasets. The results demonstrate the effectiveness of the proposed system in smart factory environments compared to existing approaches. This paper contributes to the development of lightweight intrusion detection systems for deployment on small-scale devices and provides insights into addressing the challenges of AI-based threat detection in 5G environments.