arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 0907.3754cs.CCcs.CRcs.DS
收藏

关于差分隐私的几何学

On the Geometry of Differential Privacy

Moritz Hardt, Kunal Talwar

摘要:

我们认为,在设置中,用户要求$d$线性查询$f\colon\Rn\to\Re$非自适应的噪声复杂度的差异私人机制。这里,数据库由$\Rn$中的向量表示,并且数据库之间的接近度以$\ell_1$度量来测量。我们表明,噪声的复杂性是由两个几何参数与查询集。我们使用这种连接,以提供严格的上限和下限的噪声复杂性的任何$d \leq n$。我们表明,对于$d$随机线性查询的灵敏度~1,它是必要的和充分的,以实现$ε$ -差分隐私添加$\ell_2$ -错误$Θ(\min\{d\sqrt{d}/ε,d\sqrt{\log (n/d)}/ε\})$。假设真理的深猜想从凸几何,被称为超平面猜想,我们可以将我们的结果扩展到任意的线性查询,几乎匹配的上限和下限。我们的边界转换为每个答案的错误$O(\min\{d/ε,\sqrt{d\log(n/d)}/ε\})$。最好的上界(拉普拉斯机制)给出了每个答案的$O(\min\{d/\eps,\sqrt{n}/ε\})$的界,而最知名的下界是$Ω(\sqrt{d}/ε)$。相比之下,我们的下限是强大的,足以将差分隐私的概念从近似差分隐私的概念,其中$O(\sqrt{d}/ε)$的上限可以实现。

英文摘要:

We consider the noise complexity of differentially private mechanisms in the setting where the user asks $d$ linear queries $f\colon\Rn\to\Re$ non-adaptively. Here, the database is represented by a vector in $\Rn$ and proximity between databases is measured in the $\ell_1$-metric. We show that the noise complexity is determined by two geometric parameters associated with the set of queries. We use this connection to give tight upper and lower bounds on the noise complexity for any $d \leq n$. We show that for $d$ random linear queries of sensitivity~1, it is necessary and sufficient to add $\ell_2$-error $Θ(\min\{d\sqrt{d}/ε,d\sqrt{\log (n/d)}/ε\})$ to achieve $ε$-differential privacy. Assuming the truth of a deep conjecture from convex geometry, known as the Hyperplane conjecture, we can extend our results to arbitrary linear queries giving nearly matching upper and lower bounds. Our bound translates to error $O(\min\{d/ε,\sqrt{d\log(n/d)}/ε\})$ per answer. The best previous upper bound (Laplacian mechanism) gives a bound of $O(\min\{d/\eps,\sqrt{n}/ε\})$ per answer, while the best known lower bound was $Ω(\sqrt{d}/ε)$. In contrast, our lower bound is strong enough to separate the concept of differential privacy from the notion of approximate differential privacy where an upper bound of $O(\sqrt{d}/ε)$ can be achieved.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑