论文标题
使用交互式机制，离散分布的本地私人非反应测试速度更快
Locally private non-asymptotic testing of discrete distributions is faster using interactive mechanisms
论文作者
论文摘要
我们发现在当地差异隐私的约束下测试多项式或更一般离散分布的分离率。我们构建有效的随机算法和测试程序，在这两种情况下，仅允许非相互互动的隐私机制，并且在允许所有顺序互动隐私机制的情况下。在后一种情况下，分离率更快。我们证明了一般信息的理论界限，使我们能够在大多数情况下（通常情况下）在所有的一对隐私机制和测试程序中建立算法的最佳性。被考虑的示例包括测试均匀，多项式和指数降低分布。
We find separation rates for testing multinomial or more general discrete distributions under the constraint of local differential privacy. We construct efficient randomized algorithms and test procedures, in both the case where only non-interactive privacy mechanisms are allowed and also in the case where all sequentially interactive privacy mechanisms are allowed. The separation rates are faster in the latter case. We prove general information theoretical bounds that allow us to establish the optimality of our algorithms among all pairs of privacy mechanisms and test procedures, in most usual cases. Considered examples include testing uniform, polynomially and exponentially decreasing distributions.
