Scholarlink

Federated learning allows decentralized model training while preserving data privacy. However, Non-IID data poses significant challenges, leading to performance degradation and increased communication overhead. This paper introduces FedHKD, a hierarchical FL algorithm that integrates client clustering and knowledge distillation to address Non-IID challenges. By clustering clients with similar data distributions and transferring knowledge across clusters, FedHKD improves model accuracy while reducing communication costs. Experimental results on Fashion-MNIST and CIFAR-10 datasets demonstrate up to 14.24% improvement in accuracy and a substantial reduction in communication overhead, proving its effectiveness in Non-IID environments.

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

FedHKD: A Hierarchical Federated Learning Approach Integrating Clustering and Knowledge Distillation for Non-IID Data

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Shiwen Hu,
  • Changji Wang,
  • Yuan Li,
  • Zhen Liu,
  • Ning Liu,
  • Qingqing Gan

摘要

Federated learning allows decentralized model training while preserving data privacy. However, Non-IID data poses significant challenges, leading to performance degradation and increased communication overhead. This paper introduces FedHKD, a hierarchical FL algorithm that integrates client clustering and knowledge distillation to address Non-IID challenges. By clustering clients with similar data distributions and transferring knowledge across clusters, FedHKD improves model accuracy while reducing communication costs. Experimental results on Fashion-MNIST and CIFAR-10 datasets demonstrate up to 14.24% improvement in accuracy and a substantial reduction in communication overhead, proving its effectiveness in Non-IID environments.