线性查询的一种近似最优差分隐私机制
发布时间：2018-11-07 19:02
【摘要】：在差分隐私保护程度确定的条件下使数据的有用性最大化的问题,称为差分隐私的最优机制问题.最优机制问题是差分隐私理论中的一个重要问题,与差分隐私模型的理论基础及应用前景有直接联系.与已有的研究不同,提出一种不基于敏感度的分析方法来寻找最优机制:首先,将最优机制问题构造为一个多目标函数优化问题,并提出了一种差分隐私机制构造方法,在此基础上,对线性查询问题给出了一种近似最优差分隐私机制,该机制达到了差分隐私不等式的边界.此外,大部分分析方法也可对非线性查询的最优机制问题进行分析.该研究揭示了敏感度方法的不足之处,发现其无法刻画数据集的邻居集合对应的查询函数值集合的特性,而该集合包含了差分隐私的一些深层特征.
[Abstract]:The problem of maximizing the usefulness of data under the condition that the degree of differential privacy protection is determined is called the optimal mechanism of differential privacy. The problem of optimal mechanism is an important problem in differential privacy theory, which is directly related to the theoretical basis and application prospect of differential privacy model. Different from the previous researches, an analysis method based on sensitivity is proposed to find the optimal mechanism. Firstly, the optimal mechanism problem is constructed as a multi-objective function optimization problem, and a difference privacy mechanism is proposed. On this basis, an approximate optimal differential privacy mechanism is proposed for linear query problems, which reaches the boundary of differential privacy inequality. In addition, most analysis methods can also analyze the optimal mechanism of nonlinear queries. This study reveals the inadequacies of the sensitivity method and finds that it is unable to characterize the properties of the set of query function values corresponding to the neighbor set of the dataset which contains some deep features of differential privacy.
【作者单位】： 中国科学院软件研究所基础软件国家工程研究中心;兰州财经大学信息工程学院;计算机科学国家重点实验室(中国科学院软件研究所);
【基金】：中国科学院战略性先导科技专项基金(XDA06010600)~~
【分类号】：TP309
，
本文编号：2317296
[Abstract]:The problem of maximizing the usefulness of data under the condition that the degree of differential privacy protection is determined is called the optimal mechanism of differential privacy. The problem of optimal mechanism is an important problem in differential privacy theory, which is directly related to the theoretical basis and application prospect of differential privacy model. Different from the previous researches, an analysis method based on sensitivity is proposed to find the optimal mechanism. Firstly, the optimal mechanism problem is constructed as a multi-objective function optimization problem, and a difference privacy mechanism is proposed. On this basis, an approximate optimal differential privacy mechanism is proposed for linear query problems, which reaches the boundary of differential privacy inequality. In addition, most analysis methods can also analyze the optimal mechanism of nonlinear queries. This study reveals the inadequacies of the sensitivity method and finds that it is unable to characterize the properties of the set of query function values corresponding to the neighbor set of the dataset which contains some deep features of differential privacy.
【作者单位】： 中国科学院软件研究所基础软件国家工程研究中心;兰州财经大学信息工程学院;计算机科学国家重点实验室(中国科学院软件研究所);
【基金】：中国科学院战略性先导科技专项基金(XDA06010600)~~
【分类号】：TP309
，
本文编号：2317296
本文链接：https://www.wllwen.com/kejilunwen/ruanjiangongchenglunwen/2317296.html
