Multimodal contrastive learning-based network intrusion detection

doi:10.1016/j.rineng.2026.111405

Multimodal contrastive learning-based network intrusion detection

2026-06-06

0

OA

AI

DOI：10.1016/j.rineng.2026.111405

Original

Original request for help

Save

Abstract

En 中文

• Multi-step multimodal contrastive learning stabilizes latent features under shift. • Cross-dataset adaptation validated on Edge-IIoT, IoT-NID, and CIC-IoMT datasets. • Improves recall on minority attacks (+4%) and short biflows (+3% F1,  ≤ 3 packets). • Yields a more compact latent space than transfer learning (DB 0.90 vs. 1.12). • Contrastive overhead is offline-only; inference feasible on emulated RPi4B.

Keywords:

Network intrusion detection

Contrastive learning

Representation learning

Multimodal learning

Multimodal contrastive learning-based network intrusion detection

Abstract

Journal

Organization