本地化差分隐私综述  

A Survey on Local Differential Privacy

作　　者：孙一帆 张锐[1,2] 陶杨 高碧柔 秦诗涵 安超 SUN Yifan;ZHANG Rui;TAO Yang;GAO Birou;QIN Shihan;AN Chao(SKLOIS,Institute of Information Engineering,Beijing 100085,China;School of Cyber Security,University of Chinese Academy of Sciences,Beijing 100049,China)

机构地区：[1]中国科学院信息工程研究所,信息安全国家重点实验室,北京100085 [2]中国科学院大学,网络空间安全学院,北京100049

出　　处：《数据与计算发展前沿》2023年第5期74-97,共24页Frontiers of Data & Computing

基　　金：国家自然科学基金项目(62172411,62172404,61972094,62202458)。

摘　　要：【目的】本地化差分隐私是优秀的隐私保护模型,能够在数据共享、发布的场景下对群体进行统计分析,保护个人数据隐私。本文围绕本地化差分隐私进行综述,为未来工作提供参考。【文献范围】本文调研了来自主流会议、期刊的本地化差分隐私领域的论文,并进行了总结归纳。【方法】本文以数据统计分析任务类型为线索,从基于本地化差分隐私模型的频率估计、均值估计、多维数据统计分析和机器学习4个方面展开调研。本文对相关研究进行了对比分析,对关键问题进行了总结,对现有工作的不足进行了讨论,对未来的研究方向进行了展望。【结果】本地化差分隐私模型能够在用户数据被采集、分析时,为用户个人数据隐私提供强有力的隐私保护。【局限】本文以数据统计分析任务类型为线索,未对图数据相关研究进行总结。【结论】本地化差分隐私作为一种优秀的隐私保护模型,得到学者们的关注后迅速发展,但是仍然面临着诸多问题和挑战,值得进一步研究和探索。[Objective]This paper systematically introduces local differential privacy and provides a reference for the protection of personal data privacy under data sharing and publishing.[Coverage]This paper investigates and summarizes papers from mainstream conferences and journals in the field of local differential privacy.[Methods]This paper takes the type of statistical data analysis task as a clue and conducts research based on local differential privacy from four aspects,which concludes frequency estimation,mean estimation,multidimensional data statistical analysis,and machine learning.This paper makes a comparative analysis of relevant studies,summarizes key issues,discusses the shortcomings of existing work,and looks forward to future research directions.[Results]The local differential privacy model can provide strong privacy protection for users'personal data privacy when user data is collected and analyzed.[Limitations]This paper takes the type of statistical data analysis task as a clue and does not summarize the research related to graph data.[Conclusions]Local differential privacy,as an excellent privacy-preserving model,has developed rapidly after gaining the attention of scholars.But it still faces many problems and challenges,which are worthy of further research and exploration.

关 键 词：本地化差分隐私 频率估计 均值估计 多维数据统计分析 机器学习 

分 类 号：TP309[自动化与计算机技术—计算机系统结构]