差分隐私中噪声添加与精度分析研究

兰州理工大学学报 ›› 2023, Vol. 49 ›› Issue (3): 94-103.

• 自动化技术与计算机技术 • 上一篇下一篇

差分隐私中噪声添加与精度分析研究

王骁识, 康海燕^*

北京信息科技大学信息管理学院, 北京 100192

收稿日期:2021-09-30 出版日期:2023-06-28 发布日期:2023-07-07
通讯作者: 康海燕(1971-),男,河北灵寿人,博士,教授. Email:kanghaiyan@126.com
基金资助:
国家社科基金年度项目(21BTQ079),教育部人文社科项目(20YJAZH046)

Research on noise addition and precision analysis in differential privacy

WANG Xiao-shi, KANG Hai-yan

School of Information Management, Beijing Information Science and Technology University, Beijing 100192, China

Received:2021-09-30 Online:2023-06-28 Published:2023-07-07

摘要/Abstract

摘要： 国内对于差分隐私定义以及所使用的基本机制缺乏严格清晰的证明与推导过程,对学者入门造成了困难.因此针对这方面空白,通过分析、证明与应用举例的方式,对差分隐私中的拉普拉斯机制与指数机制进行了详细分析,并给出完整的数学推导过程和应用举例.指出了拉普拉斯机制精度公式和指数机制精度公式存在缩放过大的问题,并且在拉普拉斯机制和指数机制精度公式的证明之后给出了放缩过大的理由.通过实验得出结论,拉普拉斯机制精度公式和指数机制精度公式是精度范围过大的公式.

关键词: 隐私保护, 差分隐私, 拉普拉斯机制, 指数机制

Abstract: Differential privacy, as a definition of privacy protection with a strong mathematical background and rigorous mathematical derivation, has received great attention and has been widely studied. However, up to now, there is a lack of strict and clear proof and derivation processes for the definition of differential privacy and the basic mechanism used in China, which makes it difficult for scholars to get started. Therefore, in this paper, the Laplace and exponential mechanism in differential privacy are analyzed in detail by means of analysis, proof, and application examples, and a complete mathematical derivation process and application examples are given to address this gap. It is pointed out that there is a problem of excessive scaling in the accuracy formula of the Laplace and exponential mechanism. After proving the accuracy formula of Laplace and the exponential mechanism, the reason why this paper considers excessive scaling is given. It is not difficult to conclude from experiments that the Laplace mechanism precision formula and the exponential mechanism precision formula are formulas with an excessive range of accuracy.

Key words: privacy protection, differential privacy, Laplace mechanism, exponential mechanism

中图分类号:

TN92

王骁识, 康海燕. 差分隐私中噪声添加与精度分析研究[J]. 兰州理工大学学报, 2023, 49(3): 94-103.

WANG Xiao-shi, KANG Hai-yan. Research on noise addition and precision analysis in differential privacy[J]. Journal of Lanzhou University of Technology, 2023, 49(3): 94-103.

参考文献

[1] 熊平,朱天清,王晓峰.差分隐私保护及其应用[J].计算机学报,2014,37(1):101-122.
[2] 何贤芒,王晓阳,陈华辉,等.差分隐私保护参数ε的选取研究[J].通信学报,2015,36(12):124-130.
[3] 穆海蓉,丁丽萍,宋宇宁,等.DiffPRFs:一种面向随机森林的差分隐私保护算法[J].通信学报,2016,37(9):175-182.
[4] 陈思,付安民,柯海峰,等.MCDP:基于神经网络的多集群分布式差分隐私数据发布方法[J].电子学报,2020,48(12):2297-2303.
[5] 方晨,郭渊博,王娜,等.基于生成对抗网络的差分隐私数据发布方法[J].电子学报,2020,48(10):1983-1992.
[6] 叶阿勇,孟玲玉,赵子文,等.基于预测和滑动窗口的轨迹差分隐私保护机制[J].通信学报,2020,41(4):123-133.
[7] 王斌,张磊,张国印.敏感渐进不可区分的位置隐私保护[J].计算机研究与发展,2020,57(3):160-174.
[8] 朱维军,游庆光,杨卫东,等.基于统计差分的轨迹隐私保护[J].计算机研究与发展,2017,54(12):179-186.
[9] 吴云乘,陈红,赵素云,等.一种基于时空相关性的差分隐私轨迹保护机制[J].计算机学报,2018,41(422):37-50.
[10] 霍峥,孟小峰.一种满足差分隐私的轨迹数据发布方法[J].计算机学报,2018,41(2):400-412.
[11] 杨旭东,高岭,王海,等.一种面向直方图发布的均衡差分隐私保护方法[J].计算机学报,2020,43(8):1414-1432.
[12] 张啸剑,孟小峰.基于差分隐私的流式直方图发布方法[J].软件学报,2016,27(2):381-393.
[13] 彭慧丽,金凯忠,付聪聪,等.基于序列格的隐私时序模式挖掘方法[J].电子学报,2020,48(1):153-163.
[14] 王金艳,刘陈,傅星珵,等.差分隐私的数据流关键模式挖掘方法[J].软件学报,2019,30(3):648-666.
[15] 张啸剑,王淼,孟小峰.差分隐私保护下一种精确挖掘top-k频繁模式方法[J].计算机研究与发展,2014,51(1):104-114.
[16] 叶青青,孟小峰,朱敏杰,等.本地化差分隐私研究综述[J].软件学报,2018,29(7):1981-2005.
[17] ERLINGSSON L,PIHUR V,KOROLOVA A.RAPPOR:Randomized aggregatable privacy-preserving ordinal response[J].ACM,2014(12):1054-1067.
[18] ANDREW G,CHIEN S,PAPERNOT N.Tensor flow privacy [DB/OL].[2021-06-10].https://github.com/tensorflow/privacy.
[19] DWORK C,FELDMAN V,HARDT M,et al.The reusable holdout:Preserving validity in adaptive data analysis[J].Science,2015,349(6248):636-638.
[20] SHOKRI R,SHMATIKOV V.Privacy-preserving deep learning[C]//ACM Conference on Computer and Communications Security (CCS).Monticello:IEEE,2015:1310-1321.
[21] ABADI M,CHU A,GOODFELLOW I,et al.Deep learning with differential privacy[C]//2016 ACM SIGSAC Conference.New York:ACM,2016:308-318.
[22] PAPERNOT N,ABADI M,ERLINGSSON L,et al.Semi-supervised knowledge transfer for deep learning from private training data [C/OL].[2021-06-10].https://arxiv.org/pdf/1610.05755.pdf.
[23] DONG J,ROTH A,SU W J.Gaussian differential privacy[J].Journal of the Royal Statistical Society Series B:Statistical Methodology,2022,84(1):3-37.
[24] BU Z,DONG J,LING Q,et al.Deep learning with Gaussian differential privacy [J/OL].[2021-06-10].https://www.xueshufan.com/publication/2990147108.
[25] WARNER S L.Randomized response:a survey technique for eliminating evasive answer bias[J].Publications of the American Statistical Association,1965,60(309):63-69.
[26] 张啸剑,孟小峰.面向数据发布和分析的差分隐私保护[J].计算机学报,2014,37(4):927-949.
[27] WORK C D.Differential privacy[C]//Proceedings of the 33rd international conference on Automata,Languages and Programming-Volume Part Ⅱ.Berlin,Heidelberg:Springer,2006:1-12.
[28] DWORK C,MCSHERRY F,NISSIM K,et al.Calibrating noise to sensitivity in private data analysis[J].Journal of Privacy and Confidentiality,2017,7(3):17-51.
[29] MCSHERRY F,TALWAR K.Mechanism design via differential privacy[C]//48th Annual IEEE Symposium on Foundations of Computer Science.Providence:IEEE,2007:94-103.

差分隐私中噪声添加与精度分析研究

Research on noise addition and precision analysis in differential privacy

PDF

可视化

摘要/Abstract

引用本文

使用本文

参考文献

相关文章 1

编辑推荐

Metrics

本文评价