Being able to efficiently and accurately select the top-$k$ elements with differential privacy is an integral component of various private data analysis tasks. In this paper, we present the oneshot Laplace mechanism, which generalizes the well-known Report Noisy Max mechanism to reporting noisy top-$k$ elements. We show that the oneshot Laplace mechanism with a noise level of $\widetilde{O}(\sqrt{k}/\eps)$ is approximately differentially private. Compared to the previous peeling approach of running Report Noisy Max $k$ times, the oneshot Laplace mechanism only adds noises and computes the top $k$ elements once, hence much more efficient for large $k$. In addition, our proof of privacy relies on a novel coupling technique that bypasses the use of composition theorems. Finally, we present a novel application of efficient top-$k$ selection in the classical problem of ranking from pairwise comparisons.
翻译:能够有效和准确地选择有差异隐私的顶价- k$元素是各种私人数据分析任务的一个组成部分。 在本文中, 我们展示了单张的 Laplace 机制, 将著名的《 吵闹的Max 》 报告机制概括为报告吵闹的顶价- k$元素。 我们显示, 单张的 Laplace 机制的噪音水平为 $@ lobaltilde{O} (\\ sqrt{k}/\eps), 相对以往的“ 吵闹报告 Max $ $ $ ” 的剥皮方法来说, 我们用单张Laplace 机制只增加噪音, 并一次性计算顶值的美元元素, 从而对大额的美元效率更高。 此外, 我们的隐私证据依赖于一种新颖的组合技术, 绕过组成符号的使用 。 最后, 在典型的比对比的经典问题中, 我们提出了一种新应用高效的顶价选择 $ 。