Pré-Publication, Document De Travail Année : 2025

On the MIA Vulnerability Gap Between Private GANs and Diffusion Models

Résumé

Generative Adversarial Networks (GANs) and diffusion models have emerged as leading approaches for high-quality image synthesis. While both can be trained under differential privacy (DP) to protect sensitive data, their sensitivity to membership inference attacks (MIAs), a key threat to data confidentiality, remains poorly understood. In this work, we present the first unified theoretical and empirical analysis of the privacy risks faced by differentially private generative models. We begin by showing, through a stability-based analysis, that GANs exhibit fundamentally lower sensitivity to data perturbations than diffusion models, suggesting a structural advantage in resisting MIAs. We then validate this insight with a comprehensive empirical study using a standardized MIA pipeline to evaluate privacy leakage across datasets and privacy budgets. Our results consistently reveal a marked privacy robustness gap in favor of GANs, even in strong DP regimes, highlighting that model type alone can critically shape privacy leakage.

Fichier principal
Vignette du fichier
MIA_on_DP_Models (6).pdf (262.31 Ko) Télécharger le fichier
Origine Fichiers produits par l'(les) auteur(s)
Licence

Dates et versions

hal-05236329 , version 1 (02-09-2025)

Licence

Identifiants

  • HAL Id : hal-05236329 , version 1

Citer

Ilana Sebag, Jean-Yves Franceschi, Alain Rakotomamonjy, Alexandre Allauzen, Jamal Atif. On the MIA Vulnerability Gap Between Private GANs and Diffusion Models. 2025. ⟨hal-05236329⟩
178 Consultations
238 Téléchargements

Partager

  • More