PANORAMIA: Privacy Auditing of Machine Learning Models without Retraining

TL;DR

PANORAMIA is a privacy auditing framework for machine learning models that uses membership inference attacks with generated non-member data, avoiding retraining or model modification.

Contribution

It introduces a novel privacy measurement method that does not require in-distribution non-member data or retraining, simplifying privacy audits.

Findings

Effective on image, tabular, and language models.

Does not require access to original training data.

Eliminates need for retraining during privacy assessment.

Abstract

We present PANORAMIA, a privacy leakage measurement framework for machine learning models that relies on membership inference attacks using generated data as non-members. By relying on generated non-member data, PANORAMIA eliminates the common dependency of privacy measurement tools on in-distribution non-member data. As a result, PANORAMIA does not modify the model, training data, or training process, and only requires access to a subset of the training data. We evaluate PANORAMIA on ML models for image and tabular data classification, as well as on large-scale language models.