基于茫然传输协议的FATE联邦迁移学习方案  

FATE Federated Transfer Learning Scheme Based on Oblivious Transfer Protocol

作　　者：郑云涛 叶家炜 ZHENG Yuntao;YE Jiawei(Shanghai Key Laboratory of Intelligent Information Processing,School of Computer Science,Fudan University,Shanghai 200433,China;Engineering Research Center of Cyber Security Auditing and Monitoring,Ministry of Education,School of Computer Science,Fudan University,Shanghai 200433,China)

机构地区：[1]复旦大学计算机科学技术学院,上海市智能信息处理重点实验室,上海200433 [2]复旦大学计算机科学技术学院,教育部网络信息安全审计与监控工程研究中心,上海200433

出　　处：《计算机工程》2023年第2期24-30,共7页Computer Engineering

基　　金：上海市基础研究重点项目(21JC1400600)。

摘　　要：利用不同来源的数据参与机器学习模型的训练,能够使得训练出的模型的预测结果更加准确,然而大量的数据收集则会产生隐私方面的相关问题。FATE联邦迁移学习是一种基于同态加密的联邦学习框架,但FATE联邦迁移学习中同态加密计算复杂,收敛速度相对较慢,导致模型训练效率低。提出一种基于茫然传输协议的安全矩阵计算方案。通过实现矩阵加法和乘法及数乘的安全计算,完成参与两方交互下具有数据隐私保护特性机器学习模型的损失函数计算与梯度更新,并以此构造更高效的FATE联邦迁移学习算法方案。在此基础上,通过茫然传输扩展协议和通信批量处理,减少需要调用的茫然传输协议的数量,缩减通信轮数,从而降低茫然传输协议带来的通信消耗。性能分析结果表明,该方案的安全模型满足安全性和隐私保护性,并且具有一定的可扩展性,在局域网环境下,相比基于同态加密的方案,模型收敛的平均时间缩短约25%,并且随着数据样本特征维度的增加,该方案仍能保持稳定的收敛速度。Machine learning models can be trained on a large amount of data collected from different sources,thereby improving the prediction accuracy.However,data collection between different organizations causes privacy issues.The Federated Artificial intelligence Technology Enabler(FATE) Federated Transfer Learning(FTL) is a secure machine learning framework based on homomorphic encryption.However,the convergence speed is relatively low owing to the computational efficiency of homomorphic encryption.In this study,a secure multiparty computation scheme for matrix computation based on Oblivious Transfer(OT) protocol is proposed to design a two-party machine-learning scheme to construct a more efficient FATE FTL.In addition,the communication consumption caused by the OT protocol is reduced via OT extension and batch processing.The performance analysis shows that the scheme ensures security,privacy preservation,and scalability in practical applications.On the other hand,the convergence time of the proposed scheme is approximately 25% better than that FATE FTL framework based on a homomorphic encryption scheme in a local-area network environment.As the feature dimension of the data samples increases,the advantage of the convergence speed of this scheme can remain stable,proving that this scheme has a practical application significance.

关 键 词：联邦迁移学习 安全多方计算 秘密共享 茫然传输协议 同态加密 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]