融合零知识证明的去中心联邦学习可信聚合方案

doi:10.3778/j.issn.1002-8331.2402-0066

摘要/Abstract

摘要： 去中心联邦学习可以有效解决传统联邦学习存在的通信效率低、不可信聚合服务器阻碍模型正确聚合的问题。然而，在去中心架构下，客户端成为本地训练和模型聚合过程的唯一实体，不可信客户端将直接影响联邦学习过程的完整性和模型聚合结果的正确性。基于此，利用简洁非交互零知识证明技术提出一种针对环形架构去中心联邦学习的可信聚合方案，通过对不同客户端生成的可信证据进行组合，实现客户端本地训练和模型聚合过程的正确可验证。此外，使用矩阵承诺技术确保中间参数的真实性。实验结果表明数据样本维度小于64×64时方案是可行的，且证明所需时间开销和存储开销与基准方案相比有所降低。

关键词: 隐私安全, 联邦学习, 非交互零知识证明

Abstract: Decentralized federated learning can solve the problems in typical federated learning techniques effectively, such as low communication efficiency and untrusted aggregation servers, which can prevent model aggregation from being executed correctly. However, clients become the entities of local training and model aggregation in decentralized architecture. Untrusted clients will affect the completeness of federated learning process and the correctness of aggregation result directly. This paper proposes a trustworthy aggregation scheme for the annular decentralized federated learning system based on zk-SNARK technique. The paper demonstrates the correctness and verifiability of local training and model aggregation by combining trusted proofs of different clients. Furthermore, this paper uses matrix commitment to ensure the genuineness of intermediate parameters. Experimental results show the feasibility of this scheme when the dimension of data sample is lower than 64×64. The time and storage overheads are reduced compared with the baseline scheme.

Key words: privacy security, federated learning, zero knowledge succinct non-interactive arguments of knowledge (zk-SNARK)

张首勋, 王玲玲, 耿克, 陆忠锴. 融合零知识证明的去中心联邦学习可信聚合方案[J]. 计算机工程与应用, 2025, 61(8): 274-282.

ZHANG Shouxun, WANG Lingling, GENG Ke, LU Zhongkai. Trustful Aggregation Scheme for Decentralized Federated Learning Based on Zero Knowledge Proof[J]. Computer Engineering and Applications, 2025, 61(8): 274-282.

参考文献

[1] KAIROUZ P, MCMAHAN H B, AVENT B, et al. Advances and open problems in federated learning[J]. Foundations and Trends in Machine Learning, 2021, 14(1): 1-210.
[2] ZHOU M, YANG Z, YU H, et al. VDFChain: secure and veri-fiable decentralized federated learning via committee-based blockchain[J]. Journal of Network and Computer Applications, 2024, 223: 103814.
[3] 周炜, 王超, 徐剑, 等. 基于区块链的隐私保护去中心化联邦学习模型[J]. 计算机研究与发展, 2022, 59(11): 2423-2436.
ZHOU W, WANG C, XU J, et al. Privacy-perserving and decentralized federated learning model based on the blockchain[J]. Journal of Computer Research and Development, 2022, 59(11): 2423-2436.
[4] LIU J, WANG Z, YANG Q, et al. A three-layer security assurance model for a decentralized federated learning system[C]//Proceedings of the 31st Wireless and Optical Communications Conference, 2022: 85-90.
[5] HAN G, ZHANG T, ZHANG Y, et al. Verifiable and privacy preserving federated learning without fully trusted centers[J]. Journal of Ambient Intelligence and Humanized Computing, 2021, 13(3): 1431-1441.
[6] CHE C, LI X, CHEN C, et al. A decentralized federated learning framework via committee mechanism with convergence guarantee[J]. IEEE Transactions on Parallel and Distributed Systems, 2022, 33(12): 4783-4800.
[7] XU G, LI H, LIU S, et al. VerifyNet: secure and verifiable federated learning[J]. IEEE Transactions on Information Forensics and Security, 2019, 15: 911-926.
[8] GOLDWASSER S, KALAI Y T, ROTHBLUM G N. Delegating computation: interactive proofs for muggles[J]. Journal of the ACM, 2015, 62(4): 1-64.
[9] BEN-SASSON E, CHIESA A, TROMER E, et al. Succinct non-interactive zero knowledge for a von neumann architecture[C]//Proceedings of the 23rd USENIX Security Symposium, 2014: 781-796.
[10] WAHBY R S, TZIALLA I, SHELAT A, et al. Doubly-efficient zkSNARKs without trusted setup[C]//Proceedings of the IEEE Symposium on Security and Privacy, 2018: 926-943.
[11] GUO X, LIU Z, LI J, et al. VeriFL: communication-efficient and fast verifiable aggregation for federated learning[J]. IEEE Transactions on Information Forensics and Security, 2020, 16: 1736-1751.
[12] ZHAO L, WANG Q, WANG C, et al. VeriML: enabling integrity assurances and fair payments for machine learning as a service[J]. IEEE Transactions on Parallel and Distributed Systems, 2021, 32(10): 2524-2540.
[13] LEE S, KO H, KIM J, et al. vCNN: verifiable convolutional neural network based on zk?SNARKs[J]. IEEE Transactions on Dependable and Secure Computing, 2020, 21(4): 4254-4270.
[14] FENG B, QIN L, ZHANG Z, et al. ZEN: an optimizing compiler for verifiable, zero-knowledge neural network inferences[J]. Cryptology ePrint Archive, 2021.
[15] GHODSI Z, GU T, GARG S. SafetyNets: verifiable execution of deep neural networks on an untrusted cloud[C]//Proceedings of the 31st International Conference on Neural Information Processing Systems, 2017: 4675-4684.
[16] LIU T, XIE X, ZHANG Y. zkCNN: zero knowledge proofs for convolutional neural network predictions and accuracy[C]//Proceedings of the ACM SIGSAC Conference on Computer and Communications Security, 2021: 2968-2985.
[17] JU C, LEE H, CHUNG H, et al. Efficient sum-check protocol for convolution[J]. IEEE Access, 2021, 9: 164047-164059.
[18] WENG C, YANG K, XIE X, et al. Mystique: efficient conversions for zero-knowledge proofs with applications to machine learning[C]//Proceedings of the 30th USENIX Security Symposium, 2021: 501-518.
[19] PETKUS M. Why and how zk-SNARK works[J]. arXiv:1906.07221, 2019.
[20] PARNO B, HOWELL J, GENTRY C, et al. Pinocchio: nearly practical verifiable computation[J]. Communications of the ACM, 2016, 59(2): 103-112.
[21] GROTH J. On the size of pairing-based non-interactive arguments[C]//Proceedings of the 35th Annual International Conference on the Theory and Applications of Cryptographic Techniques, 2016: 305-326.
[22] KEUFFER J, MOLVA R, CHABANNE H. Efficient proof composition for verifiable computation[C]//Proceedings of the 23rd European Symposium on Research in Computer Security, 2018: 152-171.
[23] FAN Y, XU B, ZHANG L, et al. Validating the integrity of convolutional neural network predictions based on zero-knowledge proof[J]. Information Sciences, 2023, 625: 125-140.
[24] WANG L, ZHAO X, LU Z, et al. Enhancing privacy preservation and trustworthiness for decentralized federated learning[J]. Information Sciences, 2023, 628: 449-468.
[25] LIU J, ZHANG L F. Matproofs: maintainable matrix commitment with efficient aggregation[C]//Proceedings of the ACM SIGSAC Conference on Computer and Communications Security, 2022: 2041-2054.