基于Bayes网络的高维感知数据本地隐私保护发布  

Bayesian network-based high-dimensional crowdsourced data publication with local differential privacy

作　　者：任雪斌 徐静怡 杨新宇 杨树森 Xuebin REN;Jingyi XU;Xinyu YANG;Shusen YANG(School of Computer Science and Technology,Xi'an Jiaotong University,Xi'an 710049,China;School of Mathematics and Statistics,Xi'an Jiaotong University,Xi'an 710049,China)

机构地区：[1]西安交通大学计算机科学与技术学院 [2]西安交通大学数学与统计学院

出　　处：《中国科学：信息科学》2019年第12期1586-1605,共20页Scientia Sinica(Informationis)

基　　金：国家自然科学基金(批准号:61572398,61772410,61802298,U1811461,11690011);国家重点研发计划(批准号:2017YFB1010004);中央高校基本科研业务费(批准号:xjj2018237);中国博士后基金(批准号:2017M623177)资助项目

摘　　要：群智感知系统通过对高维感知数据的发布和分析为人们带来巨大数据价值的同时,也给参与者的隐私带来了极大的隐患.目前,各种基于差分隐私的隐私保护方法被提出,但大部分方法不能同时解决高维感知数据间复杂的属性关联问题和来自不可信服务器的隐私威胁问题.基于此,本文提出了基于Bayes网络的高维感知数据本地隐私保护发布机制.该机制实现了用户端的本地数据保护,杜绝了其他方直接访问用户原始数据的可能,根本上保护了用户的数据隐私.感知服务器端在接收到用户本地隐私保护的数据后,基于Bayes网络方法对高维数据的维度相关性进行识别,将高维数据属性集划分为多个相对独立的低维属性集,进而依次合成新的数据集,可以有效地保留原始感知数据的属性维度相关性,保证合成数据集与原始数据集具有尽可能相似的统计特性.通过大量仿真实验验证了该方法的有效性,实验结果表明该方法在有效的本地隐私保护下的合成数据具有较高的数据效用性.High-dimensional crowdsourced data is pervasive in crowdsensing systems and with the development of IoTs it can produce rich knowledge for the society. However, it also creates serious privacy threats for crowdsourcing participants. To mitigate the privacy concerns in crowdsensing systems, local differential privacy has been derived from the de facto standard of differential privacy in order to achieve strong privacy guaranteed in distributed systems. However, directly achieving local differential privacy on high-dimensional crowdsourced data may lead not only to a prohibitive computational burden but also low data utility. Therefore, in this paper, we propose a local private high-dimensional data publication scheme for crowdsensing systems. In particular, on the participants’ side, high-dimensional records are locally perturbed to protect privacy, while on the server’s side, the probability distribution of original data is recovered by taking advantage of both the expectation maximization algorithm and the theory of the Bayesian network. Extensive experiments on real-world datasets demonstrated the effectiveness of the proposed scheme that can synthesize approximate datasets with local differential privacy.

关 键 词：群智感知系统 感知数据 高维数据 本地差分隐私 BAYES网络 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]