指数衰减模式下基于矩阵机制的差分隐私流数据发布算法

An algorithm for differential privacy streaming data publication based on matrix mechanism under exponential decay mode

当前许多实际应用需要持续地对流数据进行统计发布,并且对当前的数据关注度高于对历史数据的关注度.现有关于该问题的解决方案是使数据项带有权重,并提出指数衰减下的差分隐私流数据发布方法.然而,现有的方法仅考虑单次查询,未能有效利用连续统计发布背景下查询间的关联性,以进一步提高查询的精度.为此,本文利用矩阵在处理关联性查询方面的优势,提出一种指数衰减模式下基于矩阵机制的差分隐私流数据发布算法DMFDA.算法首先使用构造法生成满足流数据实时发布要求的矩阵分解策略;其次,利用对角矩阵对构造的策略矩阵进行调整,以提高发布精度;最后,根据所构造策略矩阵的子结构特性,提出快速求解对角矩阵的方法.实验对算法DMFDA发布的查询结果与同类指数衰减流数据发布算法进行比较分析.实验结果表明,算法DMFDA是有效可行的.

At present, many practical applications require the continuous release of statistical streaming data,and the importance of current data is higher than historical data. The solution to this problem is to assign weights to the data and propose a differential privacy data release method under exponential decay. However, existing methods only consider a single query, and cannot effectively use the correlation between queries in the continuous statistical publishing background to further improve the accuracy of the query. In this paper, we present a differential privacy data release algorithm（DMFDA） in exponential decay mode based on a matrix mechanism,which uses the advantages of the matrix to deal with relevant queries. Firstly, we use the construction method to generate the matrix decomposition strategy to meet the real-time requirements of streaming data. Secondly, the diagonal matrix is used to adjust the structure of the constructed strategy matrix so as to improve the release accuracy. Finally, according to the substructure of the constructed strategy matrix, a fast method of solving the diagonal matrix is proposed. The experiment is designed to compare DMFDA and similar algorithms for streaming data release in exponential decay. Experimental results show that the DMFDA algorithm is effective and feasible.