English

Debiasing Functions of Private Statistics in Postprocessing

Cryptography and Security 2025-12-17 v4 Methodology

Abstract

Given a differentially private unbiased estimate q~=q(D)+ν\tilde{q}=q(D) +\nu of a statistic q(D)q(D), we wish to obtain unbiased estimates of functions of q(D)q(D), such as 1/q(D)1/q(D), solely through post-processing of q~\tilde{q}, with no further access to the confidential dataset DD. To this end, we adapt the deconvolution method used for unbiased estimation in the statistical literature, deriving unbiased estimators for a broad family of twice-differentiable functions when the privacy-preserving noise ν\nu is drawn from the Laplace distribution (Dwork et al., 2006). We further extend this technique to a more general class of functions, deriving approximately optimal estimators that are unbiased for values in a user-specified interval (possibly extending to ±\pm \infty). We use these results to derive an unbiased estimator for private means when the size nn of the dataset is not publicly known. In a numerical application, we find that a mechanism that uses our estimator to return an unbiased sample size and mean outperforms a mechanism that instead uses the previously known unbiased privacy mechanism for such means (Kamath et al., 2023). We also apply our estimators to develop unbiased transformation mechanisms for per-record differential privacy, a privacy concept in which the privacy guarantee is a public function of a record's value (Seeman et al., 2024). Our mechanisms provide stronger privacy guarantees than those in prior work (Finley et al., 2024) by using Laplace, rather than Gaussian, noise. Finally, using a different approach, we go beyond Laplace noise by deriving unbiased estimators for polynomials under the weak condition that the noise distribution has sufficiently many moments.

Keywords

Cite

@article{arxiv.2502.13314,
  title  = {Debiasing Functions of Private Statistics in Postprocessing},
  author = {Flavio Calmon and Elbert Du and Cynthia Dwork and Brian Finley and Grigory Franguridi},
  journal= {arXiv preprint arXiv:2502.13314},
  year   = {2025}
}

Comments

This version is the same as version 2, which we inadvertently withdrew in trying to undo a premature submission. Relative to version 1, this version contains additional results and more references

R2 v1 2026-06-28T21:49:27.419Z