A Contrastive Dual-Task Framework for Few-Shot Traffic Classification in IoT Networks.

Area of Science:

• Computer Science
• Network Security
• Machine Learning

Background:

• Encrypted sensor traffic classification is vital for Internet of Things (IoT) and Mobile Edge Computing (MEC) security.
• Current methods struggle with new traffic types and distinguishing intrinsic behaviors from shared library patterns, especially under distribution shifts.

Purpose of the Study:

• To propose CDTF, a contrastive dual-task framework for transferable and few-shot traffic representation learning.
• To improve the accuracy and adaptability of encrypted traffic classification in MEC environments.

Main Methods:

• CDTF employs a hybrid pre-training strategy combining supervised triplet pretraining (STP) and self-supervised dynamic burst masking (DBM).
• STP uses base-class labels to align intra-class and separate inter-class samples, mitigating interference from shared network components.
• DBM models global semantic structures and enhances representation robustness against network noise and distribution shifts.

Main Results:

• CDTF learns discriminative and contextual representations in a shared embedding space.
• The framework enables rapid adaptation to novel categories via lightweight fine-tuning, reducing reliance on large labeled datasets.
• Experiments across nine datasets demonstrated superior performance over state-of-the-art methods, with a 4.61 percentage point Precision improvement in the few-shot setting.

Conclusions:

• CDTF offers a robust and adaptable solution for encrypted sensor traffic classification in MEC.
• The proposed framework significantly enhances few-shot learning capabilities and reduces data supervision requirements.
• CDTF demonstrates strong transferability and robustness, outperforming existing methods in diverse environments.