面向数据流的差分隐私直方图发布

Differential Private Histogram Publication for Data Stream

针对现有数据流相关的差分隐私发布技术主要应用于二进制数据流,不能高效地处理一般性数据流发布中隐私的问题,提出一种高效、面向分布不均匀的数值型数据流的差分隐私直方图发布算法——DDPA。该算法基于滑动窗口模型,利用相邻2个时间戳的数据集分布的相似性,动态合理分配隐私预算,使得每一个窗口的总预算不超过隐私预算ε,并利用分组与合并策略,快速计算出局部最优直方图。通过对该算法发布数据的可用性与同类算法进行比较分析,实验结果表明,该算法是有效可行的。

Current research on differential private publication associcted with data stream mainly considers a binary data stream, which cannot efficiently deal with the general data stream＇ s private publication. An efficient differential private histogram publication algorithm called DDPA was proposed, which is oriented toward non-uniform distributed numerical stream. Basing on the sliding window model, the similarity on two adjacent timestamps of data distribution is applied to allocate the budget privacy dynamically, which makes each window＇ s total budge not exceed the privacy budget ε, and after that, the grouping and merging strategies are used to calculate the local optimal histogram quickly. According to comparing and analyzing the proposed algorithm with the other similar algorithms on the published data＇ s availability, the experimental results show that the proposed algorithm is effective and feasible.