Scholarlink

Malicious client nodes in federated learning can corrupt the global model of central server by modifying dataset or submitted model, making it less accurate or distorted. A TD3-based malicious node detection and robust federated aggregation algorithm is proposed. The algorithm excludes malicious clients based on model distance, and uses reinforcement learning to assign model weight to client nodes to reduce the influence of malicious nodes gradually. The experimental results show that the proposed algorithm is effective against data and model modification attacks and is superior to traditional aggregation algorithms such as FedAvg, Krum and MKrum.

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

Malicious Node Detection and Robust Federated Aggregation Algorithm Based on TD3

  • 会议论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年

查看全文
  • Fan Sun,
  • Hong Wen,
  • Wenjing Hou,
  • Huanhuan Song,
  • Yongfeng Wang,
  • Ruixiang Yao,
  • Dibao Yan

摘要

Malicious client nodes in federated learning can corrupt the global model of central server by modifying dataset or submitted model, making it less accurate or distorted. A TD3-based malicious node detection and robust federated aggregation algorithm is proposed. The algorithm excludes malicious clients based on model distance, and uses reinforcement learning to assign model weight to client nodes to reduce the influence of malicious nodes gradually. The experimental results show that the proposed algorithm is effective against data and model modification attacks and is superior to traditional aggregation algorithms such as FedAvg, Krum and MKrum.