MambaTIG: Fast and Robust Encrypted Traffic Detection Leveraging Selective State-Space Masking

Summary

MambaTIG is a detection framework for encrypted network traffic that combines self-supervised masked graph learning with selective state-space modeling. It is pretrained only on benign traffic, uses a node-importance-driven masking strategy for robustness under distribution shifts, and replaces attention with a Mamba module to reduce inference overhead. Across four public datasets it reports competitive mean F1, a 56.29% average relative F1 gain under distribution-shift settings, a 37.63% adversarial-setting F1 gain, and a 66.7% reduction in average inference time.