Intrusion Detection with Federated Learning and Conditional Generative Adversarial Network in Satellite-Terrestrial Integrated Networks

Network intrusion detection is a challenging network security research topic, especially when data privacy has become an increasing concern in satellite-terrestrial integrated networks. Federated learning was introduced as an effective distributed learning scheme. However, existing studies have primarily focused on terrestrial networks. In this study, we propose a federated learning framework based on a conditional generative adversarial network (CGAN) model for intrusion detection in satellite-terrestrial integrated networks. We further propose an efficient federated learning scheme called federated learning with dynamic weight and momentum (FedDWM) for aggregating local model parameters from terrestrial clients to satellite fed servers. Numerical experiments with the CIC-IDS2017 and CSE-CIC-IDS2018 datasets demonstrate the effectiveness of the proposed approach over baselines for imbalanced intrusion detection.