Optimal differential privacy mechanisms under Hamming distortion for structured source classes
Optimal differential privacy mechanisms under Hamming distortion for structured source classes
复制标题
DOI：
10.1109/isit.2016.7541663
复制
发表时间：
2016-07
期刊：
影响因子：
--
通讯作者：
Kousha Kalantari;L. Sankar;A. Sarwate
中科院分区：
文献类型：
--
作者：
Kousha Kalantari;L. Sankar;A. Sarwate
我们研究了隐私和效用之间的权衡在局部差分隐私（L-DP）和汉明失真下的汉明失真的某些类的有限字母表源。我们定义了两类：置换不变和有序统计（其概率质量函数是单调的）。我们获得了最佳的L-DP机制的置换不变的来源，并推导出上界和下界的有序统计量的目标失真值的范围内可实现的局部差分隐私。
We examine a tradeoff between privacy and utility in terms of local differential privacy (L-DP) and Hamming distortion for certain classes of finite-alphabet sources under Hamming distortion. We define two classes: permutation-invariant, and ordered statistics (whose probability mass functions are monotonic). We obtain the optimal L-DP mechanism for permutation-invariant sources and derive upper and lower bounds on the achievable local differential privacy for ordered statistics for a range of target distortion values.