Saved in:
Bibliographic Details
Main Authors: Mahpud, Bar, Sheffet, Or
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2510.00790
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912987113062400
author Mahpud, Bar
Sheffet, Or
author_facet Mahpud, Bar
Sheffet, Or
contents We study the problem of learning exponential distributions under differential privacy. Given $n$ i.i.d.\ samples from $\mathrm{Exp}(λ)$, the goal is to privately estimate $λ$ so that the learned distribution is close in total variation distance to the truth. We present a simple pure $ε$-differentially private algorithm that avoids the classical dependence on the true value of $λ$. Our method leverages a structural property of the exponential distribution: its $(1-1/e)$-quantile equals $1/λ$, allowing us to estimate the rate parameter directly via private quantile estimation. The resulting learner is both conceptually simple and sample-efficient, achieving near-optimal guarantees. We further extend the method to Pareto distributions via a logarithmic reduction, prove nearly matching lower bounds using group privacy arguments, and show how approximate $(ε,δ)$-DP removes the need for externally supplied parameter bounds. Together, these results give the first tight characterization of exponential distribution learning under differential privacy using a simple $λ$-free approach.
format Preprint
id arxiv_https___arxiv_org_abs_2510_00790
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Differentially Private Learning of Exponential Distributions: Simple Algorithms and Tight Bounds
Mahpud, Bar
Sheffet, Or
Data Structures and Algorithms
We study the problem of learning exponential distributions under differential privacy. Given $n$ i.i.d.\ samples from $\mathrm{Exp}(λ)$, the goal is to privately estimate $λ$ so that the learned distribution is close in total variation distance to the truth. We present a simple pure $ε$-differentially private algorithm that avoids the classical dependence on the true value of $λ$. Our method leverages a structural property of the exponential distribution: its $(1-1/e)$-quantile equals $1/λ$, allowing us to estimate the rate parameter directly via private quantile estimation. The resulting learner is both conceptually simple and sample-efficient, achieving near-optimal guarantees. We further extend the method to Pareto distributions via a logarithmic reduction, prove nearly matching lower bounds using group privacy arguments, and show how approximate $(ε,δ)$-DP removes the need for externally supplied parameter bounds. Together, these results give the first tight characterization of exponential distribution learning under differential privacy using a simple $λ$-free approach.
title Differentially Private Learning of Exponential Distributions: Simple Algorithms and Tight Bounds
topic Data Structures and Algorithms
url https://arxiv.org/abs/2510.00790