Robust attacks reveal if data belongs to a set despite noise
A Simple Presentation of Robust Membership Inference Attacks
Cryptography and SecurityInformation TheoryMachine Learning
Summary
Sometimes, people want to know if a particular person's information was included in a big dataset, even if the data is mixed up or noisy. This paper explains a simple way to do this kind of membership attack, which still works well even when the data is very noisy. The authors focus on data that follows certain common patterns, like those that fit Gaussian or binary models. They also review earlier work on these attacks and how they connect to other research areas.
What this means in practice
- •For data privacy teams: Assess risks of exposing individual membership in datasets even when data is noisy or obfuscated using robust attack methods.
- •For statistical software developers: Improve tools to detect and mitigate membership inference vulnerabilities in analytic models involving Gaussian or binary data.
A survey. It maps existing work.
Authors
Renaud Gaucher, Ephraim Linder, Adam Smith, Jonathan Ullman, Lydia Zakynthinou
Abstract
In a membership inference attack, an attacker is given the data of a particular individual, and tries to determine if that individual was included in a particular dataset, given some kind of statistics derived from the dataset or a model trained on the dataset. Roughly, a membership inference attack is robust if it succeeds even when the statistics are highly noisy, regardless of how that noise is introduced. We give a simple presentation of robust membership inference attacks for mean estimation, covering both multivariate Gaussian distributions and binary product distributions. We also give a summary of the literature on robust membership inference, covering history, extensions, and connections to other research areas.