Differential Privacy with Compression - Statistics > Machine LearningReport as inadecuate

Differential Privacy with Compression - Statistics > Machine Learning - Download this document for free, or read online. Document in PDF available to download.

Abstract: This work studies formal utility and privacy guarantees for a simplemultiplicative database transformation, where the data are compressed by arandom linear or affine transformation, reducing the number of data recordssubstantially, while preserving the number of original input variables. Weprovide an analysis framework inspired by a recent concept known asdifferential privacy Dwork 06. Our goal is to show that, despite the generaldifficulty of achieving the differential privacy guarantee, it is possible topublish synthetic data that are useful for a number of common statisticallearning applications. This includes high dimensional sparse regression Zhouet al. 07, principal component analysis PCA, and other statistical measuresLiu et al. 06 based on the covariance of the initial data.

Author: Shuheng Zhou, Katrina Ligett, Larry Wasserman

Source: https://arxiv.org/


Related documents