English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

It is observed in the literature that data augmentation can significantlymitigate membership inference (MI) attack. However, in this work, we challengethis observation by proposing new MI attacks to utilize the information ofaugmented data. MI attack is widely used to measure the model's informationleakage of the training set. We establish the optimal membership inference whenthe model is trained with augmented data, which inspires us to formulate the MIattack as a set classification problem, i.e., classifying a set of augmentedinstances instead of a single data point, and design input permutationinvariant features. Empirically, we demonstrate that the proposed approachuniversally outperforms original methods when the model is trained with dataaugmentation. Even further, we show that the proposed approach can achievehigher MI attack success rates on models trained with some data augmentationthan the existing methods on models trained without data augmentation. Notably,we achieve a 70.1% MI attack success rate on CIFAR10 against a wide residualnetwork while the previous best approach only attains 61.9%. This suggests theprivacy risk of models trained with data augmentation could be largelyunderestimated.

How Does Data Augmentation Affect Privacy in Machine Learning?