一种保护数据隐私的匿名路由联邦学习框架  

An Anonymous Routing Federated Learning Framework for Data Privacy Protection

作　　者：李佳东 曾海涛 彭莉 汪晓丁[1] LI Jiadong;ZENG Haitao;PENG Li;WANG Xiaoding(College of Computer and Cyber Security,Fujian Normal University,Fuzhou 350117,China)

机构地区：[1]福建师范大学计算机与网络空间安全学院,福州350117

出　　处：《信息网络安全》2025年第3期494-503,共10页Netinfo Security

基　　金：国家自然科学基金[U1905211]。

摘　　要：联邦学习作为一种分布式机器学习框架,能够在不共享训练数据的前提下,实现多方参与者协同训练全局模型,从而有效确保客户端数据隐私安全。然而,联邦学习仍面临模型参数泄露风险和通信过程中的身份隐私威胁。针对上述问题,文章提出一种保护数据隐私的匿名路由联邦学习框架(SecFL),旨在确保联邦学习模型中的参数安全与可信传输。SecFL设计了一种组配对洋葱路由协议,基于配对的密码学对数据进行分层加密,并引入“组”的概念,使组内所有节点能够解密相应层,从而在保证消息机密性和安全性的同时提升系统匿名性。实验结果表明,SecFL在匿名路由性能与安全防护效果方面均显著优于传统方案。相较于洋葱路由和广播匿名路由,SecFL在更短时间内使消息传递率达到100%,源节点和目的节点的匿名性分别提升了3.9%和1.9%。在50%节点遭受攻击的情况下,路径匿名性指标最多提升了24.8%。此外,SecFL框架在联邦学习中的收敛性能也较好。Federated learning is a distributed machine learning framework that enables multiple participants to collaboratively train a global model without sharing their training data,thereby effectively protecting client data privacy.However,federated learning still faces risks related to model parameter leakage and identity privacy during communication.To address these issues,an anonymous routing federated learning framework for data privacy protection(SecFL)was designed,aimed at ensuring the secure and trustworthy transmission of model parameters in federated learning.SecFL introduced a novel group-pairing the onion router protocol,which used pairing cryptography to encrypt data layer by layer and incorporated the concept of“groups”,allowing all nodes within a group to decrypt the corresponding layer.This not only ensured the confidentiality and security of messages,but also enhanced system anonymity.Experimental results show that compared to the classic onion router and broadcast anonymous routing anonymous routing systems,SecFL achieves a 100%message delivery rate in a shorter time.The anonymity of the source and destination nodes is improved by up to 3.9%and 1.9%,respectively.The path anonymity can be increased by up to 24.8%when half of the nodes are compromised.Additionally,the SecFL framework demonstrates excellent convergence performance in federated learning.

关 键 词：联邦学习 洋葱路由 匿名通信 隐私保护 

分 类 号：TP309[自动化与计算机技术—计算机系统结构]