The Sample Complexity of Membership Inference and Privacy Auditing

Mahdi Haghifam, Adam Smith, Jonathan Ullman · 2025

Read The Sample Complexity of Membership Inference and Privacy Auditing by Mahdi Haghifam, Adam Smith, Jonathan Ullman (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2508-19458.txt.

Introduction

Mahdi Haghifam, Adam Smith, and Jonathan Ullman study a privacy risk in machine learning in a 2025 paper. An attacker may try to tell if one person's data was used to train a model. Most attacks use a set of fresh samples from the same source to make that call.

The authors ask how large that fresh set must be for an attack to work well. They study a clean test case where a model learns the mean of bell-curve data. Their proof shows the set may need to be far larger than the training set itself. That means common attacks may miss real risks.

Catalog record

CREATORMahdi Haghifam, Adam Smith, Jonathan Ullman
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE226 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2508-19458

Browse the full library · About the Tower of Babel