Scholarlink

<p>Federated learning (FL) is an innovative approach to machine learning that emphasizes data privacy by allowing model training on decentralized data sources. This study focuses on the application of FL using medical data, specifically chest X-ray images. We address the issue of traffic congestion between the server and clients by implementing three compression techniques: Pruning, Clustering, Quantization, and a hybrid technique that integrates these three methods which we name it (QPC). Our results demonstrate that clustering is the most effective compression technique, achieving superior upload and download speeds while preserving accuracy and significantly reducing traffic between the central server and clients. The average upload speed, download speed, and accuracy achieved were approximately 29.71 GB/s, 367.86 GB/s, and 97.53%, respectively. This research advances the field of FL by improving scalability and efficiency, offering valuable insights for future research in Federated Learning and healthcare applications.</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

Optimizing Federated Learning for Medical Images Analysis Through Traffic Efficiency

  • 期刊论文

  • 发表日期：

    2025年8月21日

查看全文
  • Lamees Rababa,
  • Omar AlZoubi,
  • Muneer Bani Yassein

摘要

Federated learning (FL) is an innovative approach to machine learning that emphasizes data privacy by allowing model training on decentralized data sources. This study focuses on the application of FL using medical data, specifically chest X-ray images. We address the issue of traffic congestion between the server and clients by implementing three compression techniques: Pruning, Clustering, Quantization, and a hybrid technique that integrates these three methods which we name it (QPC). Our results demonstrate that clustering is the most effective compression technique, achieving superior upload and download speeds while preserving accuracy and significantly reducing traffic between the central server and clients. The average upload speed, download speed, and accuracy achieved were approximately 29.71 GB/s, 367.86 GB/s, and 97.53%, respectively. This research advances the field of FL by improving scalability and efficiency, offering valuable insights for future research in Federated Learning and healthcare applications.