Reversible Token-Based OOD Detection and Representation Learning for Encrypted Network Traffic

​

Encrypted network traffic has become the dominant form of Internet communication, making traffic analysis increasingly challenging due to the limited availability of payload information. This research aims to develop a representation learning framework for encrypted network traffic based on reversible token representations. By converting packet sequences into reversible token representations, the proposed approach preserves structural information while enabling efficient feature learning. The study further investigates unknown traffic detection, allowing the model to identify previously unseen applications or communication patterns without relying on predefined labels. Representation learning techniques, including self-supervised learning and contrastive learning, are explored to improve feature quality and generalization across diverse network environments. The ultimate goal is to establish a robust and scalable framework for encrypted traffic analysis that supports network security, anomaly detection, and intelligent traffic management.