Scholarlink

<p>The rapid proliferation of the Internet of Medical Things (IoMT) has significantly advanced real-time healthcare monitoring and decision-making capabilities. However, large-scale IoMT deployments generate high-frequency telemetry streams and distributed security events that must be processed under strict latency and privacy constraints. Addressing these challenges requires distributed parallel processing, scalable model aggregation, and real-time network orchestration—core characteristics of modern high-performance and supercomputing systems. This paper proposes a high-performance distributed IoMT edge security framework integrating federated small language models (FSLMs) with knowledge-defined networks (KDNs). Lightweight Transformer-based models (TinyBERT and DistilBERT) are adapted for structured telemetry sequences and deployed across edge nodes, enabling privacy-preserving semantic anomaly detection. Federated aggregation with differential privacy and semantic filtering ensures robust distributed training, while the KDN control plane performs parallel telemetry analytics and dynamic policy reconfiguration through programmable data planes. Experimental evaluation using the CIC IoMT Dataset 2024 and a real-world hospital deployment demonstrates 95.1% detection accuracy, 11.4% reduction in packet loss, 5.2% throughput improvement, 40% reduction in telemetry overhead, and sub-430&#xa0;ms mitigation latency under distributed load. Scalability experiments show stable convergence across up to 200 edge devices with bounded communication complexity. These results demonstrate that secure IoMT edge intelligence inherently requires distributed learning, parallel telemetry processing, and real-time orchestration—aligning the proposed framework with the scope of high-performance and supercomputing systems.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhancing IoMT edge security through federated small language models and knowledge-defined networks

  • 期刊论文

  • 发表日期：

    2026年6月24日

查看全文
  • Ahmed Raoof Tawfeeq Al-Hasani,
  • Ali Broumandnia,
  • Hamid Haj Seyyed Javadi

摘要

The rapid proliferation of the Internet of Medical Things (IoMT) has significantly advanced real-time healthcare monitoring and decision-making capabilities. However, large-scale IoMT deployments generate high-frequency telemetry streams and distributed security events that must be processed under strict latency and privacy constraints. Addressing these challenges requires distributed parallel processing, scalable model aggregation, and real-time network orchestration—core characteristics of modern high-performance and supercomputing systems. This paper proposes a high-performance distributed IoMT edge security framework integrating federated small language models (FSLMs) with knowledge-defined networks (KDNs). Lightweight Transformer-based models (TinyBERT and DistilBERT) are adapted for structured telemetry sequences and deployed across edge nodes, enabling privacy-preserving semantic anomaly detection. Federated aggregation with differential privacy and semantic filtering ensures robust distributed training, while the KDN control plane performs parallel telemetry analytics and dynamic policy reconfiguration through programmable data planes. Experimental evaluation using the CIC IoMT Dataset 2024 and a real-world hospital deployment demonstrates 95.1% detection accuracy, 11.4% reduction in packet loss, 5.2% throughput improvement, 40% reduction in telemetry overhead, and sub-430 ms mitigation latency under distributed load. Scalability experiments show stable convergence across up to 200 edge devices with bounded communication complexity. These results demonstrate that secure IoMT edge intelligence inherently requires distributed learning, parallel telemetry processing, and real-time orchestration—aligning the proposed framework with the scope of high-performance and supercomputing systems.