PAC Privacy Preserving Diffusion Models

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Xu, Qipan, Ding, Youlong, Zhang, Xinxi, Gao, Jie, Wang, Hao
Format: Preprint
Veröffentlicht: 2023
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866912146891210752
author Xu, Qipan
Ding, Youlong
Zhang, Xinxi
Gao, Jie
Wang, Hao
author_facet Xu, Qipan
Ding, Youlong
Zhang, Xinxi
Gao, Jie
Wang, Hao
contents Data privacy protection is garnering increased attention among researchers. Diffusion models (DMs), particularly with strict differential privacy, can potentially produce images with both high privacy and visual quality. However, challenges arise such as in ensuring robust protection in privatizing specific data attributes, areas where current models often fall short. To address these challenges, we introduce the PAC Privacy Preserving Diffusion Model, a model leverages diffusion principles and ensure Probably Approximately Correct (PAC) privacy. We enhance privacy protection by integrating a private classifier guidance into the Langevin Sampling Process. Additionally, recognizing the gap in measuring the privacy of models, we have developed a novel metric to gauge privacy levels. Our model, assessed with this new metric and supported by Gaussian matrix computations for the PAC bound, has shown superior performance in privacy protection over existing leading private generative models according to benchmark tests.
format Preprint
id arxiv_https___arxiv_org_abs_2312_01201
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle PAC Privacy Preserving Diffusion Models
Xu, Qipan
Ding, Youlong
Zhang, Xinxi
Gao, Jie
Wang, Hao
Machine Learning
Artificial Intelligence
Data privacy protection is garnering increased attention among researchers. Diffusion models (DMs), particularly with strict differential privacy, can potentially produce images with both high privacy and visual quality. However, challenges arise such as in ensuring robust protection in privatizing specific data attributes, areas where current models often fall short. To address these challenges, we introduce the PAC Privacy Preserving Diffusion Model, a model leverages diffusion principles and ensure Probably Approximately Correct (PAC) privacy. We enhance privacy protection by integrating a private classifier guidance into the Langevin Sampling Process. Additionally, recognizing the gap in measuring the privacy of models, we have developed a novel metric to gauge privacy levels. Our model, assessed with this new metric and supported by Gaussian matrix computations for the PAC bound, has shown superior performance in privacy protection over existing leading private generative models according to benchmark tests.
title PAC Privacy Preserving Diffusion Models
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2312.01201