Headline result: 100+ application classes; retrieval-based adaptation
Over 90% of internet traffic is now encrypted. While encryption protects privacy, it makes traditional network monitoring impossible. We build AI systems that classify and analyze encrypted traffic without decryption — enabling security, quality of service, and compliance while preserving user privacy.
Key Contributions
- PQClass: Post-Quantum Traffic Classification — The first system to classify traffic encrypted with post-quantum algorithms (CRYSTALS-Kyber, NTRU), ensuring network visibility during the quantum cryptography transition.
- GAN-Based Data Synthesis — Generative adversarial networks that produce realistic encrypted traffic datasets, addressing the scarcity of shareable labeled data in this domain.
- Spectral Analysis for Protocol Identification — Frequency-domain analysis of packet inter-arrival times and sizes reveals application fingerprints inside TLS 1.3 and ESNI without any payload inspection.
- Zero-Day Attack Detection — SimCSE-based contrastive learning builds traffic representations that generalize to unseen attack patterns without retraining on labeled attack data.
Impact
Our methods are validated on real-world datasets from commercial ISPs and academic network captures. Applications include real-time threat detection in enterprise networks, QoE optimization for streaming services, and network capacity planning — all with full encryption preserved.
Key publications & resources
- Publications — filter by the Encrypted Traffic topic to see this project's papers.
- Code & software — lab repositories, datasets, and tools.
Collaborate with us
We welcome academic and industry collaborations, and motivated graduate students, on this research line. Get in touch →