计算机应用研究2025，Vol.42Issue(11)：3468-3475,8.DOI:10.19734/j.issn.1001-3695.2025.03.0097
基于动态群签名的区块链联邦学习隐私保护方案
Privacy-preserving scheme for blockchain-based federated learning using dynamic group signatures
摘要
Abstract
Blockchain-based federated learning is a decentralized machine learning approach that enables distributed clients to collaboratively train models.However,existing blockchain-based federated learning systems lacked adequate protection for sensitive attributes such as user identities,making collusion attacks between malicious trainers and validators easy to execute.This paper proposed a blockchain-based federated learning privacy protection scheme(DGS-BCFL)that innovatively integra-ted dynamic group signatures with blockchain federated learning.The scheme firstly utilized the anonymity of dynamic group signatures to protect user identity privacy while enabling traceability of malicious users' anonymous identities and revocation of low-contribution malicious participants.Secondly,it developed a contribution-based adaptive incentive algorithm to ensure fairness by rewarding nodes according to their workload intensity and functional roles.Finally,it evaluated the performance of DGS-BCFL on the FEMNIST dataset.Experimental results show that when facing collusion attacks from malicious nodes,DGS-BCFL achieves an 89.36％success rate in resisting such attacks,representing a 24.92％improvement over VBFL.The scheme also demonstrates 26.13％higher model accuracy compared to BDFL.Therefore,the proposed scheme not only main-tains the high robustness of BCFL,but also demonstrates superior performance in model accuracy.关键词
区块链/联邦学习/动态群签名/合谋攻击/自适应激励算法Key words
blockchain/federated learning/dynamic group signature/conspiracy attack/adaptive incentive algorithm分类
计算机与自动化引用本文复制引用
张蒙,古春生,张言,史培中,景征骏..基于动态群签名的区块链联邦学习隐私保护方案[J].计算机应用研究,2025,42(11):3468-3475,8.基金项目
国家自然科学基金资助项目(61672270,61602216) （61672270,61602216）
