Joint Semantic Transfer Network for IoT Intrusion Detection

In this article, we propose a joint semantic transfer network (JSTN) toward effective intrusion detection (ID) for large-scale scarcely labeled Internet of Things (IoT) domain. As multisource heterogeneous domain adaptation (MS-HDA) method, the JSTN integrates knowledge-rich (NI) and another small-scale IoT (II) as source domains preserves intrinsic properties to assist target II ID. The jointly transfers following three semantics learn domain-invariant discriminative feature representation. scenario endows NI with characteristics from each other ease knowledge process via confused discriminator categorical distribution preservation. It also reduces source–target discrepancy make shared space invariant. Meanwhile, weighted implicit boosts discriminability fine-grained preservation, which divergence guides importance weighting during preservation reflect degree learning. Additionally, hierarchical explicit alignment performs centroid-level representative-level help geometric similarity-aware pseudo-label refiner, exploits value unlabeled explicitly aligns representations global local perspective in concentrated manner. Comprehensive experiments on various tasks verify superiority against state-of-the-art comparing methods, average 10.3% accuracy boost is achieved. statistical soundness constituting component computational efficiency verified.