Robust Semi-Supervised Temporal Intrusion Detection for Adversarial Cloud Networks

The paper proposes a robust semi-supervised temporal learning framework for cloud network intrusion detection, using consistency regularization, confidence-aware pseudo-labeling, and selective temporal invariance to cope with adversarial contamination and temporal drift in unlabeled traffic. In limited-label evaluations on CIC-IDS2017, CSE-CIC-IDS2018, and UNSW-NB15, its detection performance, label efficiency, and resilience to adversarial and non-stationary traffic are superior to state-of-the-art supervised and semi-supervised intrusion detection systems.