Paper restates robust membership inference for mean estimation
Original title: A Simple Presentation of Robust Membership Inference Attacks
The attacks remain effective when released statistics are highly noisy, for Gaussians and binary products.
The paper gives a simplified presentation of robust membership inference attacks against mean estimation. An attacker who already has one person's data tries to tell whether that person was in a dataset from noisy statistics or from a model trained on it. The treatment covers multivariate Gaussian distributions and binary product distributions. It also summarizes prior work on robust membership inference, including history, extensions, and links to other areas.
- Robustness means success despite highly noisy statistics, whatever the noise source.
- The core setting is mean estimation rather than a named deployed model.
- The paper is a presentation and literature summary, not a new incident report.
Impact: Infer whether a known person was in a dataset
From the original article
1 min read at arxiv.orgIn 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.
By Renaud Gaucher · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.