Scholarlink

<p>Federated learning (FL) systems rely on a central server for model aggregation, introducing challenges such as synchronization issues, dishonest servers, curious clients, and participant heterogeneity. To address these challenges, we propose a secure FL framework that integrates Blockchain technology and lattice-based cryptography. Our system employs NTRU (<i>N</i>th Degree Truncated Polynomial Ring Units)-based encryption to protect model gradients during transmission, providing robust defense against passive and quantum attacks. Additionally, we leverage Blockchain technology to enable decentralized and asynchronous aggregation, ensuring data integrity and reducing the risks associated with centralized control. We use selective parameter encryption to minimize computational overhead by focusing on the most privacy-sensitive gradients, enhancing both efficiency and security. Through comprehensive simulations and theoretical analysis, we demonstrate that our framework offers strong privacy guarantees and resilience against various adversarial threats. Compared to traditional cryptographic methods such as RSA and Elliptic Curve Diffie–Hellman, our solution provides a practical and scalable approach that maintains model accuracy while safeguarding data, proving it as a robust solution for real-world FL deployments.</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

Advanced federated learning security: NTRU and blockchain synergy

  • 期刊论文

  • 发表日期：

    2025年3月19日

查看全文
  • Ashlesha Hota,
  • Arishmita Biswas,
  • Sanchita Saha,
  • Amitava Nag,
  • Ferdous Ahmed Barbhuiya,
  • Sukumar Nandi

摘要

Federated learning (FL) systems rely on a central server for model aggregation, introducing challenges such as synchronization issues, dishonest servers, curious clients, and participant heterogeneity. To address these challenges, we propose a secure FL framework that integrates Blockchain technology and lattice-based cryptography. Our system employs NTRU (Nth Degree Truncated Polynomial Ring Units)-based encryption to protect model gradients during transmission, providing robust defense against passive and quantum attacks. Additionally, we leverage Blockchain technology to enable decentralized and asynchronous aggregation, ensuring data integrity and reducing the risks associated with centralized control. We use selective parameter encryption to minimize computational overhead by focusing on the most privacy-sensitive gradients, enhancing both efficiency and security. Through comprehensive simulations and theoretical analysis, we demonstrate that our framework offers strong privacy guarantees and resilience against various adversarial threats. Compared to traditional cryptographic methods such as RSA and Elliptic Curve Diffie–Hellman, our solution provides a practical and scalable approach that maintains model accuracy while safeguarding data, proving it as a robust solution for real-world FL deployments.