English

Locally Private Sampling with Public Data

Machine Learning 2025-05-05 v2

Abstract

Local differential privacy (LDP) is increasingly employed in privacy-preserving machine learning to protect user data before sharing it with an untrusted aggregator. Most LDP methods assume that users possess only a single data record, which is a significant limitation since users often gather extensive datasets (e.g., images, text, time-series data) and frequently have access to public datasets. To address this limitation, we propose a locally private sampling framework that leverages both the private and public datasets of each user. Specifically, we assume each user has two distributions: pp and qq that represent their private dataset and the public dataset, respectively. The objective is to design a mechanism that generates a private sample approximating pp while simultaneously preserving qq. We frame this objective as a minimax optimization problem using ff-divergence as the utility measure. We fully characterize the minimax optimal mechanisms for general ff-divergences provided that pp and qq are discrete distributions. Remarkably, we demonstrate that this optimal mechanism is universal across all ff-divergences. Experiments validate the effectiveness of our minimax optimal sampler compared to the state-of-the-art locally private sampler.

Keywords

Cite

@article{arxiv.2411.08791,
  title  = {Locally Private Sampling with Public Data},
  author = {Behnoosh Zamanlooy and Mario Diaz and Shahab Asoodeh},
  journal= {arXiv preprint arXiv:2411.08791},
  year   = {2025}
}
R2 v1 2026-06-28T19:58:36.685Z