Scholarlink

<p>Vehicular networking technology using federated learning enhances data privacy and security compared to centralized methods. Yet, it requires further refinement to combat single-point failure and membership inference attacks, privacy concerns, and communication expenses. This paper addresses these challenges by integrating federated differential privacy with blockchain, introducing a reputation mechanism, and using triple-gradient techniques along with model compression to reduce communication costs. In differential privacy experiments, we have determined that federated differential privacy protection is closer to the accuracy of a no-privacy protection scheme compared to traditional differential privacy protection, especially when C ≥ 2. In ternary gradient experiments, we observed a reduction in training gradients of 14.99 ×, 15.54 ×, and 15.97 × across three datasets. In layer sensitivity experiments, we found that the accuracy at top = 97%, 94%, and 91% is comparable to that at top = 100% (uncompressed). In the blockchain experiment, we proposed a new consensus mechanism—Proof of Reputation and Learning. The results show that this mechanism effectively incentivizes vehicle nodes and eliminates malicious attackers, ensuring that the attackers' reputation values fall well below the reputation threshold.</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

Fast-CFLB: a privacy-preserving data sharing system for Internet of Vehicles using ternary federated learning and blockchain

  • 期刊论文

  • 发表日期：

    2025年8月19日

查看全文
  • Jiaheng Li,
  • Qinmu Wu

摘要

Vehicular networking technology using federated learning enhances data privacy and security compared to centralized methods. Yet, it requires further refinement to combat single-point failure and membership inference attacks, privacy concerns, and communication expenses. This paper addresses these challenges by integrating federated differential privacy with blockchain, introducing a reputation mechanism, and using triple-gradient techniques along with model compression to reduce communication costs. In differential privacy experiments, we have determined that federated differential privacy protection is closer to the accuracy of a no-privacy protection scheme compared to traditional differential privacy protection, especially when C ≥ 2. In ternary gradient experiments, we observed a reduction in training gradients of 14.99 ×, 15.54 ×, and 15.97 × across three datasets. In layer sensitivity experiments, we found that the accuracy at top = 97%, 94%, and 91% is comparable to that at top = 100% (uncompressed). In the blockchain experiment, we proposed a new consensus mechanism—Proof of Reputation and Learning. The results show that this mechanism effectively incentivizes vehicle nodes and eliminates malicious attackers, ensuring that the attackers' reputation values fall well below the reputation threshold.