20 20

Transactions on
Data Privacy
Foundations and Technologies

http://www.tdp.cat


Articles in Press

Accepted articles here

Latest Issues

Year 2018

Volume 11 Issue 3
Volume 11 Issue 2
Volume 11 Issue 1

Year 2017

Volume 10 Issue 3
Volume 10 Issue 2
Volume 10 Issue 1

Year 2016

Volume 9 Issue 3
Volume 9 Issue 2
Volume 9 Issue 1

Year 2015

Volume 8 Issue 3
Volume 8 Issue 2
Volume 8 Issue 1

Year 2014

Volume 7 Issue 3
Volume 7 Issue 2
Volume 7 Issue 1

Year 2013

Volume 6 Issue 3
Volume 6 Issue 2
Volume 6 Issue 1

Year 2012

Volume 5 Issue 3
Volume 5 Issue 2
Volume 5 Issue 1

Year 2011

Volume 4 Issue 3
Volume 4 Issue 2
Volume 4 Issue 1

Year 2010

Volume 3 Issue 3
Volume 3 Issue 2
Volume 3 Issue 1

Year 2009

Volume 2 Issue 3
Volume 2 Issue 2
Volume 2 Issue 1

Year 2008

Volume 1 Issue 3
Volume 1 Issue 2
Volume 1 Issue 1


Volume 11 Issue 1


Membership Inference Attack against Differentially Private Deep Learning Model

Md Atiqur Rahman(a),(*), Tanzila Rahman(b), Robert Laganière(a), Noman Mohammed(b)

Transactions on Data Privacy 11:1 (2018) 61 - 79

Abstract, PDF

(a) School of Electrical Engineering and Computer Science, University of Ottawa, ON, K1N 6N5, Canada.

(b) Department of Computer Science, University of Manitoba, MB, R3T2N2, Canada.

e-mail:mrahm021 @uottawa.ca; rahmant4 @myumanitoba.ca; laganier @eecs.uottawa.ca; noman @cs.umanitoba.ca


Abstract

The unprecedented success of deep learning is largely dependent on the availability of massive amount of training data. In many cases, these data are crowd-sourced and may contain sensitive and confidential information, therefore, pose privacy concerns. As a result, privacy-preserving deep learning has been gaining increasing focus nowadays. One of the promising approaches for privacy-preserving deep learning is to employ differential privacy during model training which aims to prevent the leakage of sensitive information about the training data via the trained model. While these models are considered to be immune to privacy attacks, with the advent of recent and sophisticated attack models, it is not clear how well these models trade-off utility for privacy. In this paper, we systematically study the impact of a sophisticated machine learning based privacy attack called the membership inference attack against a state-of-the-art differentially private deep model. More specifically, given a differentially private deep model with its associated utility, we investigate how much we can infer about the model's training data. Our experimental results show that differentially private deep models may keep their promise to provide privacy protection against strong adversaries by only offering poor model utility, while exhibit moderate vulnerability to the membership inference attack when they offer an acceptable utility. For evaluating our experiments, we use the CIFAR-10 and MNIST datasets and the corresponding classification tasks.

* Corresponding author.

Follow us




Supports




ISSN: 1888-5063; ISSN (Digital): 2013-1631; D.L.:B-11873-2008; Web Site: http://www.tdp.cat/
Contact: Transactions on Data Privacy; Vicenç Torra; U. of Skövde; PO Box 408; 54128 Skövde; (Sweden); e-mail:tdp@tdp.cat
Note: TDP's web site does not use cookies. TDP does not keep information neither on IP addresses nor browsers. For the privacy policy access here.

 


Vicenç Torra, Last modified: 05 : 55 April 25 2018.