交互式框架中多次查询的差分隐私算法  

Differential Privacy Algorithm for Multi-Query in Interactive Framework

作　　者：欧阳恒 龙士工[1,2] 弓晓锋 OUYANG Heng;LONG Shigong;GONG Xiaofeng(College of Computer Science and Technology,Guizhou University,Guiyang 550025;State Key Laboratory of Public Big Data,Guizhou University,Guiyang 550025;Guizhou Science and Technology Information Center,Guiyang 550025)

机构地区：[1]贵州大学计算机科学与技术学院,贵阳550025 [2]贵州大学公共大数据国家重点实验室,贵阳550025 [3]贵州省科技信息中心,贵阳550025

出　　处：《计算机与数字工程》2024年第8期2259-2265,2322,共8页Computer & Digital Engineering

基　　金：国家自然科学基金项目“面向高维数据的差分隐私保护方法及其博弈模型研究”(编号:62062020);贵州省科技厅自然科学项目“SM3国产密码算法在微服务JWT用户认证体系中的研究与应用”(编号:黔科合基础[2019]1425号)资助。

摘　　要：在差分隐私交互式框架中,数据集通常需要回答多个查询,随着隐私预算逐渐被消耗,增加了隐私泄露的风险。所以,节约和跟踪隐私预算的消耗情况至关重要,该消耗不应超过隐私预算给定的限额。针对以上问题,论文设计了多次查询的差分隐私保护机制(Multi-Query-Differential Privacy Mechanism,MQDPM),采用重用噪声的思想。相同类型的查询可以重用噪声,节约隐私预算的开销,提高数据集可支持的查询数量;其次,由于差分隐私的可用性和安全性的矛盾关系,MQDPM采用可用性更好的分析高斯机制(AGM)作为噪声扰动机制,并使用Newton下山法代替二分法迭代,降低了AGM迭代的时间复杂度。最后,利用区块链记录隐私预算,方便跟踪隐私预算的使用情况,并保证不超过给定的限额。在公开的IPUMS数据集上实验表明,论文提出的MQDPM与现有算法相比,在降低查询响应时间的同时,相同隐私预算限额下,有效地节省了隐私预算开销,并且数据可用性更高。In the differential privacy interactive framework,data sets usually need to answer multiple queries.With the gradual consumption of privacy budget,the risk of privacy disclosure is increased.Therefore,it is very important to save and track the consumption of privacy budget,which should not exceed the limit given by privacy budget.To solve the above problems,this paper designs a Multi-Query-Differential Privacy Mechanism(MQDPM).The idea of reusing noise is adopted.The same type of queries can reuse noise,save the cost of privacy budget,and improve the number of queries that can be supported by the dataset.Secondly,due to the contradiction between the availability and security of differential privacy,MQDPM uses the better availability Analysis Gaussian Mechanism(AGM)as the noise disturbance mechanism,and uses the Newton downhill method to replace the dichotomy iteration,which reduces the time complexity of AGM iteration.Finally,the blockchain is used to record the privacy budget,which is convenient to track the use of the privacy budget and ensure that it does not exceed the given limit.Experiments on the public IPUMS data set show that compared with the existing algorithms,MQDPM proposed in this paper not only reduces the query response time,but also effectively saves the privacy budget overhead under the same privacy budget limit,and has higher data availability.

关 键 词：差分隐私 交互式框架 高斯机制 隐私预算 区块链 

分 类 号：TP309[自动化与计算机技术—计算机系统结构]