Reproduce “Differentially Private Nonparametric Confidence Intervals Under Minimal Distributional Assumptions” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Tomer Shoham, Moshe Shenfeld, Noa Velner-Harris, Katrina Ligett
Researchers from The Hebrew University of Jerusalem developed PrivSub, a general framework for constructing differentially private nonparametric confidence intervals (CIs) that operates under minimal distributional assumptions. The method empirically achieved near-nominal coverage and competitive CI widths across various statistical tasks, outperforming existing general-purpose private CI methods in finite-sample settings.
Key insights
- Differentially private nonparametric CIs can be constructed by combining a full dataset estimate with multiple subsampled private estimates, leveraging privacy amplification to manage the privacy-accuracy trade-off effectively.
- The framework provides asymptotic validity and tightness for CIs under minimal distributional assumptions, requiring only the existence of a limiting distribution and a known convergence rate, without assuming asymptotic normality or smoothness of the functional.
- By centering and rescaling an empirical CDF derived from private subsample estimates, the method can accurately approximate the sampling distribution of a private statistic, offering a modular 'black-box' approach.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.