Scholarlink

<p>Federated learning is a key technology for collaborative model training between multiple clients, enabling data-driven decision-making without requiring centralized data storage. Despite its potential, traditional Federated learning systems face major challenges related to data privacy, communication efficiency, and security. To address these challenges, this study proposes a novel Federated Learning framework that integrates blockchain, fog nodes, Cheon-Kim-Kim-Song (<i>CKKS</i>) homomorphic encryption, and dynamic client selection. Blockchain guarantees secure and tamper-proof aggregation of model updates, while fog nodes enable efficient intermediate aggregation, thereby reducing communication overhead, and smart contracts are used to perform the final aggregation. The <i>CKKS</i> scheme ensures robust privacy protection by allowing computations on encrypted data, and dynamic client selection improves model accuracy by prioritizing high-quality contributions while filtering out unreliable clients. Experiments on the MNIST and CIFAR-10 datasets show that the proposed framework not only improves model accuracy and communication efficiency compared to previous privacy-oriented approaches while reducing training time, but also provides a scalable, safe and efficient solution for privacy-critical applications, which is ideally suited for real-world deployment across many domains.</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

A privacy-centric federated learning framework with blockchain and dynamic client selection mechanisms

  • 期刊论文

  • 发表日期：

    2025年10月17日

查看全文
  • Ahmed Saidi,
  • Abdelouahab Amira,
  • Omar Nouali

摘要

Federated learning is a key technology for collaborative model training between multiple clients, enabling data-driven decision-making without requiring centralized data storage. Despite its potential, traditional Federated learning systems face major challenges related to data privacy, communication efficiency, and security. To address these challenges, this study proposes a novel Federated Learning framework that integrates blockchain, fog nodes, Cheon-Kim-Kim-Song (CKKS) homomorphic encryption, and dynamic client selection. Blockchain guarantees secure and tamper-proof aggregation of model updates, while fog nodes enable efficient intermediate aggregation, thereby reducing communication overhead, and smart contracts are used to perform the final aggregation. The CKKS scheme ensures robust privacy protection by allowing computations on encrypted data, and dynamic client selection improves model accuracy by prioritizing high-quality contributions while filtering out unreliable clients. Experiments on the MNIST and CIFAR-10 datasets show that the proposed framework not only improves model accuracy and communication efficiency compared to previous privacy-oriented approaches while reducing training time, but also provides a scalable, safe and efficient solution for privacy-critical applications, which is ideally suited for real-world deployment across many domains.