arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2406.04868cs.LGcs.CRcs.DS
收藏

扰动与投影:差分隐私相似度与边际查询

Perturb-and-Project: Differentially Private Similarities and Marginals

Google Research(谷歌研究院);Bocconi(博科尼大学)

Vincent Cohen-Addad, Tommaso d'Orsi, Alessandro Epasto, Vahab Mirrokni, Peilin Zhong

AI总结:

本文在差分隐私输入扰动框架下提出高效算法,用于隐私发布余弦相似度和计算任意k路边际查询,并扩展至稀疏数据集,同时从理论上解释了快速扰动算法有效的原因。

摘要:

我们重新审视差分隐私中的输入扰动框架,在该框架下,噪声被添加到输入 $A\in \mathcal{S}$ 中,随后将结果投影回可容许数据集空间 $\mathcal{S}$。借助这一框架,我们首先设计了新颖的高效算法,用于隐私发布成对余弦相似度。其次,我们推导出一种新颖的算法,用于在 $n$ 个特征上计算 $k$ 路边际查询。此前的工作仅在 $k$ 为偶数时才能获得可比的保证。此外,我们将结果扩展到 $t$-稀疏数据集,当 $t\le n^{5/6}/\log n$ 时,我们的高效算法能够给出新颖且更强的保证。最后,我们提供了一个理论视角,解释为什么快速的输入扰动算法在实践中表现良好。我们结果背后的关键技术要素是紧的平方和证书,它们为解集的高斯复杂度提供了上界。

英文摘要:

We revisit the input perturbations framework for differential privacy where noise is added to the input $A\in \mathcal{S}$ and the result is then projected back to the space of admissible datasets $\mathcal{S}$. Through this framework, we first design novel efficient algorithms to privately release pair-wise cosine similarities. Second, we derive a novel algorithm to compute $k$-way marginal queries over $n$ features. Prior work could achieve comparable guarantees only for $k$ even. Furthermore, we extend our results to $t$-sparse datasets, where our efficient algorithms yields novel, stronger guarantees whenever $t\le n^{5/6}/\log n\,.$ Finally, we provide a theoretical perspective on why \textit{fast} input perturbation algorithms works well in practice. The key technical ingredients behind our results are tight sum-of-squares certificates upper bounding the Gaussian complexity of sets of solutions.

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。

↑