Scholarlink

<p>With the rapid development of machine learning (ML) technologies, Federated Learning (FL), as a privacy-preserving distributed learning approach, has been widely applied in fields such as healthcare, finance, and the IoT. However, a major challenge in the application of FL remains how to simultaneously protect user privacy and prevent model parameter leakage during the aggregation and training processes. Currently, Homomorphic Encryption (HE) and Trusted Execution Environment (TEE) are key technologies for building Privacy-preserving Federated Learning (PPFL) solutions. However, mainstream FL methods based on single-key HE cannot resist collusion attacks between users and the server, while those based on multi-key HE (MKHE) require that users remain online during the FL process. To address this issue, this paper proposes a PPFL based on MKHE and TEE. In this scheme, the server uses TEE to uniformly generate the public and private keys for all users and ensures secure key distribution through quantum key distribution, allowing the decryption process to proceed smoothly even if users are offline, thus avoiding decryption failures. The proposed scheme not only effectively prevents collusion attacks among users and between users and the server, but also ensures the computational and aggregation capabilities of the FL system when users are offline, thereby enhancing the robustness and security of the system. Experimental evaluation results show that the proposed scheme can maintain high computational efficiency in large-scale data environments while ensuring privacy protection.</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

Pivacy-preserving federated learning based on multi-key fully homomorphic encryption and trusted execution environment

  • 期刊论文

  • 发表日期：

    2025年5月9日

查看全文
  • Gang Liu,
  • Zheng He,
  • Le Cheng,
  • Yi Luo,
  • Senmiao Su,
  • Jingchen Su,
  • Keming Zhang

摘要

With the rapid development of machine learning (ML) technologies, Federated Learning (FL), as a privacy-preserving distributed learning approach, has been widely applied in fields such as healthcare, finance, and the IoT. However, a major challenge in the application of FL remains how to simultaneously protect user privacy and prevent model parameter leakage during the aggregation and training processes. Currently, Homomorphic Encryption (HE) and Trusted Execution Environment (TEE) are key technologies for building Privacy-preserving Federated Learning (PPFL) solutions. However, mainstream FL methods based on single-key HE cannot resist collusion attacks between users and the server, while those based on multi-key HE (MKHE) require that users remain online during the FL process. To address this issue, this paper proposes a PPFL based on MKHE and TEE. In this scheme, the server uses TEE to uniformly generate the public and private keys for all users and ensures secure key distribution through quantum key distribution, allowing the decryption process to proceed smoothly even if users are offline, thus avoiding decryption failures. The proposed scheme not only effectively prevents collusion attacks among users and between users and the server, but also ensures the computational and aggregation capabilities of the FL system when users are offline, thereby enhancing the robustness and security of the system. Experimental evaluation results show that the proposed scheme can maintain high computational efficiency in large-scale data environments while ensuring privacy protection.