ja4

This paper explores applying JEPA-style predictive learning to JA4-derived network fingerprints, building a Transformer-based model (JA4-JEPA) trained on JA4, JA4H, JA4S, and JA4X subfields. The model achieves strong performance on protocol-family classification, suggesting JEPA objectives can work for compact network fingerprint representations.