Tight Auditing of Differential Privacy in MST and AIM

TL;DR

This paper introduces a Gaussian Differential Privacy-based auditing framework that provides tight privacy guarantees for state-of-the-art DP synthetic data generators MST and AIM, revealing a small gap between theory and practice.

Contribution

The paper presents the first tight auditing method for MST and AIM under strong privacy settings using GDP, enabling precise privacy measurement.

Findings

Achieved a close match between empirical and theoretical privacy measures (0.43 vs. 0.45) at (,\u03b4)=(1,10^{-2})

Provided the first tight audits in the strong-privacy regime for these generators

Code is publicly available at https://github.com/sassoftware/dpmm

Abstract

State-of-the-art Differentially Private (DP) synthetic data generators such as MST and AIM are widely used, yet tightly auditing their privacy guarantees remains challenging. We introduce a Gaussian Differential Privacy (GDP)-based auditing framework that measures privacy via the full false-positive/false-negative tradeoff. Applied to MST and AIM under worst-case settings, our method provides the first tight audits in the strong-privacy regime. For $(\epsilon,\delta)=(1,10^{-2})$, we obtain $\mu_{emp}\approx0.43$ vs. implied $\mu=0.45$, showing a small theory-practice gap. Our code is publicly available: https://github.com/sassoftware/dpmm.