联邦学习隐私保护机制综述  

Survey on Privacy-Preserving Mechanism in Federated Learning

作　　者：王浩竣 梁亚楠 黎琳[1] 李锐 Wang Haojun;Liang Yanan;Li Lin;Li Rui(College of Computer and Information Technology,Beijing Jiaotong University,Beijing 100091)

机构地区：[1]北京交通大学计算机与信息技术学院,北京100091

出　　处：《现代计算机》2022年第14期1-12,共12页Modern Computer

摘　　要：随着数据孤岛的出现和隐私意识的增强,传统的中心化的机器学习模式遇到了一系列挑战。联邦学习作为一种新兴的隐私保护的分布式机器学习模型迅速成为一个热门的研究问题。有研究表明,机器学习模型的梯度会泄露用户数据集的隐私,能够被攻击者利用以获取非法的利益,因此,需要采用一些隐私保护的机制来保护这种敏感信息。本文研究了当前联邦学习系统中采用的隐私保护机制,并根据研究者采用的隐私保护技术,将联邦学习中的隐私保护机制分为五类,总结了不同的隐私保护机制的研究思路和研究进展。通过对当前联邦学习中使用的隐私保护机制的研究,联邦学习系统的设计人员可以提高联邦学习系统的安全性,更好地保护数据隐私。With the emergence of data islands and the enhancement of privacy awareness,the traditional centralized ma⁃chine learning model has encountered a series of challenges.Federated learning,as a privacy-protected distributed machine learn⁃ing model,has quickly become a hot research topic.Studies have shown that the gradient of the machine learning model will leak the privacy of user data sets and can be used by attackers to obtain illegal benefits.Therefore,some privacy protection mechanisms are needed to protect this sensitive information.We study the privacy protection mechanisms used in the current federal learning system,and according to the privacy protection technology adopted by the researchers,divides the privacy protection mechanisms in the federal learning into five categories,and summarizes the ideas of different privacy protection mechanisms.We want to intro⁃duce the current privacy protection mechanisms used in federated learning to enable designers of federated systems to improve the security of federated learning systems and protect data privacy.

关 键 词：联邦学习 隐私保护 密码学 

分 类 号：TP309[自动化与计算机技术—计算机系统结构]