Membership inference attack against principal component analysis

Zari, Oualid; Parra-Arnau, Javier; Unsal, Ayse; Strufe, Thorsten; Önen, Melek
PSD 2022, Privacy in Statistical Databases, 21-23 September 2022, Paris, France

This paper studies the performance of membership inference attacks against principal component analysis (PCA). In this attack, we assume that the adversary has access to the principal components, and her main goal is to infer whether a given data sample was used to compute these principal components. We show that our attack is successful and achieves high performance when the number of samples used to compute the principal components is small. As a defense strategy, we investigate the use of various differentially private mechanisms. Accordingly, we present experimental results on the performance of Gaussian and Laplace mechanisms under naive and advanced compositions against
MIA as well as the utility of these differentially-private PCA solutions.

DOI
HAL
Type:
Conférence
City:
Paris
Date:
2022-09-21
Department:
Sécurité numérique
Eurecom Ref:
7041
Copyright:
© Springer. Personal use of this material is permitted. The definitive version of this paper was published in PSD 2022, Privacy in Statistical Databases, 21-23 September 2022, Paris, France and is available at : https://doi.org/10.1007/978-3-031-13945-1_19

PERMALINK : https://www.eurecom.fr/publication/7041