Differentially Private Generative Adversarial Networks with Model Inversion

Dongjie Chen; Sen Ching Samson Cheung; Chen Nee Chuah; Sally Ozonoff

doi:10.1109/WIFS53200.2021.9648378

Differentially Private Generative Adversarial Networks with Model Inversion

Dongjie Chen, Sen Ching Samson Cheung, Chen Nee Chuah, Sally Ozonoff

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

2 Scopus citations

Abstract

To protect sensitive data in training a Generative Adversarial Network (GAN), the standard approach is to use differentially private (DP) stochastic gradient descent method in which controlled noise is added to the gradients. The quality of the output synthetic samples can be adversely affected and the training of the network may not even converge in the presence of these noises. We propose Differentially Private Model Inversion (DPMI) method where the private data is first mapped to the latent space via a public generator, followed by a lower-dimensional DP-GAN with better convergent properties. Experimental results on standard datasets CIFAR10 and SVHN as well as on a facial landmark dataset for Autism screening show that our approach outperforms the standard DP-GAN method based on Inception Score, Frechet Inception Distance, and classification accuracy under the same privacy guarantee.

Original language	English
Title of host publication	2021 IEEE International Workshop on Information Forensics and Security, WIFS 2021
ISBN (Electronic)	9781665417174
DOIs	https://doi.org/10.1109/WIFS53200.2021.9648378
State	Published - 2021
Event	2021 IEEE International Workshop on Information Forensics and Security, WIFS 2021 - Montpellier, France Duration: Dec 7 2021 → Dec 10 2021

Publication series

Name	2021 IEEE International Workshop on Information Forensics and Security, WIFS 2021

Conference

Conference	2021 IEEE International Workshop on Information Forensics and Security, WIFS 2021
Country/Territory	France
City	Montpellier
Period	12/7/21 → 12/10/21

Bibliographical note

Funding Information:
VI. ACKNOWLEDGEMENTS Research reported in this publication was supported by the National Institutes of Health, United States of America under award number R01MH121344-01 and the Child Family Endowed Professorship.

Publisher Copyright:
© 2021 IEEE.

Keywords

Generative adversarial networks
differential privacy
model inversion

ASJC Scopus subject areas

Computer Networks and Communications
Information Systems
Information Systems and Management
Safety, Risk, Reliability and Quality

Access to Document

10.1109/WIFS53200.2021.9648378

Cite this

@inproceedings{0449a472f0ea4fe989a7ac0ef78a1386,

title = "Differentially Private Generative Adversarial Networks with Model Inversion",

abstract = "To protect sensitive data in training a Generative Adversarial Network (GAN), the standard approach is to use differentially private (DP) stochastic gradient descent method in which controlled noise is added to the gradients. The quality of the output synthetic samples can be adversely affected and the training of the network may not even converge in the presence of these noises. We propose Differentially Private Model Inversion (DPMI) method where the private data is first mapped to the latent space via a public generator, followed by a lower-dimensional DP-GAN with better convergent properties. Experimental results on standard datasets CIFAR10 and SVHN as well as on a facial landmark dataset for Autism screening show that our approach outperforms the standard DP-GAN method based on Inception Score, Frechet Inception Distance, and classification accuracy under the same privacy guarantee.",

keywords = "Generative adversarial networks, differential privacy, model inversion",

author = "Dongjie Chen and Cheung, {Sen Ching Samson} and Chuah, {Chen Nee} and Sally Ozonoff",

note = "Funding Information: VI. ACKNOWLEDGEMENTS Research reported in this publication was supported by the National Institutes of Health, United States of America under award number R01MH121344-01 and the Child Family Endowed Professorship. Publisher Copyright: {\textcopyright} 2021 IEEE.; 2021 IEEE International Workshop on Information Forensics and Security, WIFS 2021 ; Conference date: 07-12-2021 Through 10-12-2021",

year = "2021",

doi = "10.1109/WIFS53200.2021.9648378",

language = "English",

series = "2021 IEEE International Workshop on Information Forensics and Security, WIFS 2021",

booktitle = "2021 IEEE International Workshop on Information Forensics and Security, WIFS 2021",

}

Chen, D, Cheung, SCS, Chuah, CN & Ozonoff, S 2021, Differentially Private Generative Adversarial Networks with Model Inversion. in 2021 IEEE International Workshop on Information Forensics and Security, WIFS 2021. 2021 IEEE International Workshop on Information Forensics and Security, WIFS 2021, 2021 IEEE International Workshop on Information Forensics and Security, WIFS 2021, Montpellier, France, 12/7/21. https://doi.org/10.1109/WIFS53200.2021.9648378

Differentially Private Generative Adversarial Networks with Model Inversion. / Chen, Dongjie; Cheung, Sen Ching Samson; Chuah, Chen Nee et al.
2021 IEEE International Workshop on Information Forensics and Security, WIFS 2021. 2021. (2021 IEEE International Workshop on Information Forensics and Security, WIFS 2021).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - Differentially Private Generative Adversarial Networks with Model Inversion

AU - Chen, Dongjie

AU - Cheung, Sen Ching Samson

AU - Chuah, Chen Nee

AU - Ozonoff, Sally

N1 - Funding Information: VI. ACKNOWLEDGEMENTS Research reported in this publication was supported by the National Institutes of Health, United States of America under award number R01MH121344-01 and the Child Family Endowed Professorship. Publisher Copyright: © 2021 IEEE.

PY - 2021

Y1 - 2021

N2 - To protect sensitive data in training a Generative Adversarial Network (GAN), the standard approach is to use differentially private (DP) stochastic gradient descent method in which controlled noise is added to the gradients. The quality of the output synthetic samples can be adversely affected and the training of the network may not even converge in the presence of these noises. We propose Differentially Private Model Inversion (DPMI) method where the private data is first mapped to the latent space via a public generator, followed by a lower-dimensional DP-GAN with better convergent properties. Experimental results on standard datasets CIFAR10 and SVHN as well as on a facial landmark dataset for Autism screening show that our approach outperforms the standard DP-GAN method based on Inception Score, Frechet Inception Distance, and classification accuracy under the same privacy guarantee.

AB - To protect sensitive data in training a Generative Adversarial Network (GAN), the standard approach is to use differentially private (DP) stochastic gradient descent method in which controlled noise is added to the gradients. The quality of the output synthetic samples can be adversely affected and the training of the network may not even converge in the presence of these noises. We propose Differentially Private Model Inversion (DPMI) method where the private data is first mapped to the latent space via a public generator, followed by a lower-dimensional DP-GAN with better convergent properties. Experimental results on standard datasets CIFAR10 and SVHN as well as on a facial landmark dataset for Autism screening show that our approach outperforms the standard DP-GAN method based on Inception Score, Frechet Inception Distance, and classification accuracy under the same privacy guarantee.

KW - Generative adversarial networks

KW - differential privacy

KW - model inversion

UR - http://www.scopus.com/inward/record.url?scp=85124132323&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85124132323&partnerID=8YFLogxK

U2 - 10.1109/WIFS53200.2021.9648378

DO - 10.1109/WIFS53200.2021.9648378

M3 - Conference contribution

AN - SCOPUS:85124132323

T3 - 2021 IEEE International Workshop on Information Forensics and Security, WIFS 2021

BT - 2021 IEEE International Workshop on Information Forensics and Security, WIFS 2021

T2 - 2021 IEEE International Workshop on Information Forensics and Security, WIFS 2021

Y2 - 7 December 2021 through 10 December 2021

ER -

Differentially Private Generative Adversarial Networks with Model Inversion

Abstract

Publication series

Conference

Bibliographical note

Keywords

ASJC Scopus subject areas

Access to Document

Other files and links

Fingerprint

Cite this