Rényi Differential Privacy of the Sampled Gaussian Mechanism

Ilya Mironov, Kunal Talwar, Li Zhang
8/28/2019
cs.LGcs.CRstat.ML

Official code: tensorflow/privacy (2,040 stars, Apache-2.0, last commit 2026-10-12).

Abstract

The Sampled Gaussian Mechanism (SGM)---a composition of subsampling and the additive Gaussian noise---has been successfully used in a number of machine learning applications. The mechanism's unexpected power is derived from privacy amplification by sampling where the privacy cost of a single evaluation diminishes quadratically, rather than linearly, with the sampling rate. Characterizing the precise privacy properties of SGM motivated development of several relaxations of the notion of differential privacy. This work unifies and fills in gaps in published results on SGM. We describe a numerically stable procedure for precise computation of SGM's Rényi Differential Privacy and prove a nearly tight (within a small constant factor) closed-form bound.

View on arXivView PDF

Code Implementations(1)

tensorflow/privacyOfficial100%
Apache-2.0

Library for training machine learning models with privacy for training data

2,040477Shell, PythonDec 21, 20181 weeks agoApache-2.0
machine-learningprivacy

Similar Papers

Enhancing Trade-offs in Privacy, Utility, and Computational Efficiency through MUltistage Sampling Technique (MUST)

89% match2023

Unified Mechanism-Specific Amplification by Subsampling and Group Privacy Amplification

87% match1 repo32024

Bounding data reconstruction attacks with the hypothesis testing interpretation of differential privacy

86% match2023

Privacy Accounting for Empirical Denoisers

86% match2026

Faster Privacy Accounting via Evolving Discretization

85% match2022

On Choosing the μ Parameter in Gaussian Differential Privacy

85% match2026

Cite this paper

@article{mironov2019rnyi,
  title  = {Rényi Differential Privacy of the Sampled Gaussian Mechanism},
  author = {Ilya Mironov and Kunal Talwar and Li Zhang},
  year   = {2019},
  eprint = {1908.10530},
  archivePrefix = {arXiv},
  url    = {https://arxiv.org/abs/1908.10530}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/1908.10530.svg)](https://codeofpaper.com/paper/1908.10530)

Discussion