How Does Data Augmentation Affect Privacy in Machine Learning?

TL;DR

This paper introduces a novel set classification approach for membership inference attacks that reveals data augmentation can increase privacy risks in machine learning models, contrary to prior beliefs.

Contribution

The work formulates MI attack as a set classification problem with permutation-invariant features, demonstrating improved attack success rates on models trained with data augmentation.

Findings

Proposed method outperforms existing MI attack techniques.

Achieved 70.1% attack success rate on CIFAR10, surpassing previous methods.

Data augmentation may significantly elevate privacy risks in ML models.

Abstract

It is observed in the literature that data augmentation can significantly mitigate membership inference (MI) attack. However, in this work, we challenge this observation by proposing new MI attacks to utilize the information of augmented data. MI attack is widely used to measure the model's information leakage of the training set. We establish the optimal membership inference when the model is trained with augmented data, which inspires us to formulate the MI attack as a set classification problem, i.e., classifying a set of augmented instances instead of a single data point, and design input permutation invariant features. Empirically, we demonstrate that the proposed approach universally outperforms original methods when the model is trained with data augmentation. Even further, we show that the proposed approach can achieve higher MI attack success rates on models trained with some data augmentation than the existing methods on models trained without data augmentation. Notably, we achieve a 70.1% MI attack success rate on CIFAR10 against a wide residual network while the previous best approach only attains 61.9%. This suggests the privacy risk of models trained with data augmentation could be largely underestimated.