Differentially Private Confidence Intervals for Empirical Risk Minimization

Wang, Yue; Kifer, Daniel; Lee, Jae Woo

doi:10.29012/jpc.660

Cited by 26 publications

(25 citation statements)

References 28 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Few studies address differentially private confidence intervals and hypothesis tests specifically for regression coefficients. Wang et al (2018) developed confidence intervals based on Chaudhuri et al's (2011) techniques with satisfactory empirical coverage probability for data sets with independent and identically distributed records. However, these rely on asymptotic normality of the gradient of the objective function, which may not be satisfied for sample sizes smaller than those used in the study's empirical evaluation, where n was 30 000 or more.…”

Section: Discussionmentioning

confidence: 99%

“…Unlike the techniques of Sheffet (2017), these do not require boundedness assumptions on the underlying data. For more general regression models, Wang et al (2018) developed confidence intervals satisfying zero‐concentrated differential privacy for coefficients produced by the regularised methods of Chaudhuri et al (2011). They exploit a large‐sample normal approximation to the gradient of the objective function at its optimum, under the assumption that data set rows are randomly sampled.…”

Section: Confidence Intervals and Hypothesis Testsmentioning

confidence: 99%

See 1 more Smart Citation

A Survey of Differentially Private Regression for Clinical and Epidemiological Research

Ficek

Wang

Chen

et al. 2020

Int Statistical Rev

View full text Add to dashboard Cite

Summary Differential privacy is a framework for data analysis that provides rigorous privacy protections for database participants. It has increasingly been accepted as the gold standard for privacy in the analytics industry, yet there are few techniques suitable for statistical inference in the health sciences. This is notably the case for regression, one of the most widely used modelling tools in clinical and epidemiological studies. This paper provides an overview of differential privacy and surveys the literature on differentially private regression, highlighting the techniques that hold the most relevance for statistical inference as practiced in clinical and epidemiological research. Research gaps and opportunities for further inquiry are identified.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Confidence Intervals and Hypothesis Testsmentioning

confidence: 99%

A Survey of Differentially Private Regression for Clinical and Epidemiological Research

Ficek

Wang

Chen

et al. 2020

Int Statistical Rev

View full text Add to dashboard Cite

show abstract

“…More recently, Duchi et al (2018) release differentially private estimators for means, medians, generalized linear models, and non-parametric densities. Wang et al (2019) generate confidence intervals for parameters obtained from differentially private empirical risk minimization machine learning models. Other researchers found that differential privacy-part of our goal-was too strong a requirement for meaningful data utility on highly detailed data sets (Abowd et al 2013;Bambauer et al 2013); however, recent research is more promising.…”

Section: Literature Reviewmentioning

confidence: 99%

Protecting survey data on a consumer level

Schneider

Iacobucci

2020

J Market Anal

View full text Add to dashboard Cite

This paper offers an easy-to-implement approach to protect multivariate survey data common in marketing, such as attitudes and demographics. Our approach preserves multivariate distributions by releasing a protected data set with privacy protections. The data represent a highly detailed multivariate survey with severe privacy issues that enables us to demonstrate the tradeoff between data utility and data privacy. We create a data privacy metric that quantifies the ability of a data intruder successfully identify survey respondents and their sensitive responses. We provide data privacy measurements for a variety of competitor methods such as sampling and random noise addition and we show that by comparison, our approach can prevent a data intruder from targeting individuals while maintaining a very high level of data utility.

show abstract

“…Previous work in [21,48,58] has investigated how to accurately compute the test statistics in hypothesis testing while using DP to protect data. Ding et al designed an algorithm to detect privacy violations of DP mechanisms from a hypothesis testing perspective [12].…”

Section: Hypothesis Testing In Dpmentioning

confidence: 99%

Investigating Statistical Privacy Frameworks from the Perspective of Hypothesis Testing

Liu

Chanyaswad³

et al. 2019

Proceedings on Privacy Enhancing Technologies

View full text Add to dashboard Cite

Over the last decade, differential privacy (DP) has emerged as the gold standard of a rigorous and provable privacy framework. However, there are very few practical guidelines on how to apply differential privacy in practice, and a key challenge is how to set an appropriate value for the privacy parameter ɛ. In this work, we employ a statistical tool called hypothesis testing for discovering useful and interpretable guidelines for the state-of-the-art privacy-preserving frameworks. We formalize and implement hypothesis testing in terms of an adversary’s capability to infer mutually exclusive sensitive information about the input data (such as whether an individual has participated or not) from the output of the privacy-preserving mechanism. We quantify the success of the hypothesis testing using the precision- recall-relation, which provides an interpretable and natural guideline for practitioners and researchers on selecting ɛ. Our key results include a quantitative analysis of how hypothesis testing can guide the choice of the privacy parameter ɛ in an interpretable manner for a differentially private mechanism and its variants. Importantly, our findings show that an adversary’s auxiliary information - in the form of prior distribution of the database and correlation across records and time - indeed influences the proper choice of ɛ. Finally, we also show how the perspective of hypothesis testing can provide useful insights on the relationships among a broad range of privacy frameworks including differential privacy, Pufferfish privacy, Blowfish privacy, dependent differential privacy, inferential privacy, membership privacy and mutual-information based differential privacy.

show abstract

Differentially Private Confidence Intervals for Empirical Risk Minimization

Cited by 26 publications

References 28 publications

A Survey of Differentially Private Regression for Clinical and Epidemiological Research

A Survey of Differentially Private Regression for Clinical and Epidemiological Research

Protecting survey data on a consumer level

Investigating Statistical Privacy Frameworks from the Perspective of Hypothesis Testing

Contact Info

Product

Resources

About