arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2006.09352cs.DScs.CRcs.LGstat.ML
收藏

适用于大多数机器学习任务的一次性私人草图

A One-Pass Private Sketch for Most Machine Learning Tasks

Rice University(莱斯大学)

Benjamin Coleman, Anshumali Shrivastava

摘要:

差分隐私(DP)是一个令人信服的隐私定义,通过正式的,可证明的保证解释了隐私效用权衡。受通用数据发布算法的最新进展的启发,我们提出了一个私人草图,或数据集的小摘要,支持多种机器学习任务,包括回归,分类,密度估计,近邻搜索等。我们的草图由随机化的列联表组成,这些列联表使用位置敏感的散列进行索引,并使用高效的一次通过算法构建。我们证明了DP核密度估计的竞争性误差界。DP核密度估计的现有方法规模很差,通常随着维度的增加而呈指数增长。相比之下,我们的sketch可以在一个通道中快速运行大型高维数据集。详尽的实验表明,我们的通用草图提供了一个类似的隐私效用权衡相比,现有的DP方法的计算成本的一小部分。我们希望我们的草图能够在分布式的大规模机器学习环境中实现差分隐私。

英文摘要:

Differential privacy (DP) is a compelling privacy definition that explains the privacy-utility tradeoff via formal, provable guarantees. Inspired by recent progress toward general-purpose data release algorithms, we propose a private sketch, or small summary of the dataset, that supports a multitude of machine learning tasks including regression, classification, density estimation, near-neighbor search, and more. Our sketch consists of randomized contingency tables that are indexed with locality-sensitive hashing and constructed with an efficient one-pass algorithm. We prove competitive error bounds for DP kernel density estimation. Existing methods for DP kernel density estimation scale poorly, often exponentially slower with an increase in dimensions. In contrast, our sketch can quickly run on large, high-dimensional datasets in a single pass. Exhaustive experiments show that our generic sketch delivers a similar privacy-utility tradeoff when compared to existing DP methods at a fraction of the computation cost. We expect that our sketch will enable differential privacy in distributed, large-scale machine learning settings.

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。

↑