One-Shot Private Confidence Regions via Resampling

2026-10-06

AI Summary

This study addresses the high computational and privacy overhead caused by sequential noise addition in differentially private confidence interval construction by proposing a single-noise resampling framework. The method adds noise exclusively to the final quantile, thereby avoiding the privatization of intermediate statistics. By integrating Gaussian differential privacy, m-out-of-n resampling, and smooth sensitivity analysis, it reduces the privacy cost to logarithmic order (with replacement) or constant order (without replacement) relative to the number of resamples, eliminating the square-root factor inherent in traditional approaches. The theoretical analysis provides non-asymptotic privacy and utility guarantees applicable to mean-type estimators and degenerate U-statistics, significantly reducing private error and enhancing practical utility.

Abstract

We propose a simple framework for constructing differentially private confidence regions \textit{in one shot}, i.e., by adding noise only to the final resampling quantile instead of privatizing the estimator computed on each resample. The cost of privacy of our procedure is only logarithmic in the number of resamples $B$ under with-replacement ($m$-out-of-$n$) sampling and independent of $B$ under without replacement sampling (subsampling), avoiding the $\sqrt{B}$ factor that arises in previous works. We provide nonasymptotic Gaussian Differential Privacy (GDP) and utility guarantees for both subsampling and $m$-out-of-$n$ resampling, covering mean-like estimators with small global sensitivity as well as estimators admitting efficiently computable smooth sensitivity bounds, including quantiles and degenerate U-statistics. This allows us to also obtain private confidence regions for degenerate U-statistics where the private error is much smaller than the non-private error. In all, we provide a toolbox for widely applicable DP uncertainty quantification procedures under popular resampling strategies while avoiding the computational and privacy costs of privatizing many intermediate resample statistics.

Problem

• Differential Privacy
• Confidence Regions
• Resampling
• Uncertainty Quantification
• Privacy Cost

Innovation

• Differential Privacy
• One-Shot Resampling
• Confidence Regions
• Gaussian Differential Privacy
• Degenerate U-statistics