Scholarlink

Federated learning excels in network attack detection through global model aggregation and distribution. To tackle client data heterogeneity, we propose pFedKC, a knowledge-complementary personalized federated learning approach. First, GMM models client datasets, and data distribution information is uploaded to the server to measure heterogeneity and identify complementary terminals. Global and complementary losses guide local model training, and server-side model aggregation uses cosine similarity weights. Experiments with varying heterogeneity levels demonstrate that terminal knowledge complementation enhances heterogeneous terminal models’ attack detection capabilities.

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

Network Attack Detection Technology Based on Knowledge Complementary Personalized Federated Learning

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Shuanghui Wan,
  • Yanping Xu,
  • Ming Xu,
  • Zeqi He,
  • Dedong Zhang,
  • Yongxing Xu,
  • Yanbo Fang,
  • Hua Zhang,
  • Yifan Wu

摘要

Federated learning excels in network attack detection through global model aggregation and distribution. To tackle client data heterogeneity, we propose pFedKC, a knowledge-complementary personalized federated learning approach. First, GMM models client datasets, and data distribution information is uploaded to the server to measure heterogeneity and identify complementary terminals. Global and complementary losses guide local model training, and server-side model aggregation uses cosine similarity weights. Experiments with varying heterogeneity levels demonstrate that terminal knowledge complementation enhances heterogeneous terminal models’ attack detection capabilities.