Abstract: Federated Learning enables collaborative model training while preserving data privacy. However, challenges remain in privacy protection, participant trust, and defense against adversarial attacks. In Hierarchical Federated Learning , untrusted central servers, intermediaries, and edge devices pose risks of data leakage and malicious manipulation. Additionally, adversarial clients may upload abnormal gradients, compromising model performance. Efficient security verification and adversarial detection in HFL are therefore critical issues. To address the challenges of mutual distrust among participants and Byzantine attacks in hierarchical federated learning, a secure aggregation scheme with non-interactive verification under hierarchical architecture is proposed. First, a mutual verification mechanism for hierarchical federated learning is designed based on a commitment scheme, allowing participants to perform mutual verification. Second, a constraint and detection scheme for malicious updates is constructed using non-interactive zero-knowledge range proofs, enabling the server to detect and exclude malicious users. Third, a noise masking scheme is designed based on the Chinese Remainder Theorem, supporting user exit and reconnection while ensuring local user privacy. Finally, security analysis and experimental evaluation demonstrate that the proposed scheme can achieve secure mutual verification and malicious detection with high efficiency.

摘要： 联邦学习作为一种分布式机器学习框架，能够在保护数据隐私的同时实现模型协同训练。然而，联邦学习在隐私保护、参与方互信及恶意攻击等方面仍面临挑战，尤其是在分层架构联邦学习中，中心服务器、中间层及终端设备的不可信性可能导致隐私泄露或恶意操纵。此外，恶意用户可能上传异常参数破坏训练进程，影响模型性能。因此，如何在分层联邦学习中高效实现安全验证和恶意检测，成为亟待解决的问题。本文针对分层架构联邦学习中的参与方无法互信、拜占庭攻击等问题，提出一种分层架构下的非交互式验证联邦学习安全聚合方案。首先，基于承诺方案设计多层架构下的联邦学习非交互式验证机制，允许各参与方进行互相验证。其次，基于零知识范围证明构造恶意更新的约束与检测方案，使服务器能检测并剔除恶意用户。再次，基于中国剩余定理设计噪声掩码方案，在保证用户本地隐私的同时，还支持用户的退出与重连；最后，安全性分析与实验评估表明，本方案能够以较高的效率实现安全相互验证以及恶意检测。