Scholarlink

With the continuous expansion of the Internet of Things (IoT) network, efficient distributed anomaly detection is very important. This paper focuses on improving the Federated Principal Component Analysis (FedPG) algorithm based on Grassmann manifold, and solves the problem of slow convergence caused by its fixed learning rate by introducing an adaptive learning rate mechanism. The improved algorithm (EFedPCA) can dynamically adjust the learning rate, thus accelerating the convergence. Experiments on UNSW-NB15 and TON-IoT data sets show that EFedPCA has significantly improved convergence speed and detection accuracy compared with the original FedPG algorithm.

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

IoT Anomaly Detection Based on Grassmann Manifold Federated PCA

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Yiying Zhang,
  • Xiaokun Wang,
  • Jiajie Sun,
  • Huifeng Bai,
  • Huadong Yu,
  • Rongxu Hou

摘要

With the continuous expansion of the Internet of Things (IoT) network, efficient distributed anomaly detection is very important. This paper focuses on improving the Federated Principal Component Analysis (FedPG) algorithm based on Grassmann manifold, and solves the problem of slow convergence caused by its fixed learning rate by introducing an adaptive learning rate mechanism. The improved algorithm (EFedPCA) can dynamically adjust the learning rate, thus accelerating the convergence. Experiments on UNSW-NB15 and TON-IoT data sets show that EFedPCA has significantly improved convergence speed and detection accuracy compared with the original FedPG algorithm.