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

Privacy-preserving machine learning aims to train models on private datawithout leaking sensitive information. Differential privacy (DP) is consideredthe gold standard framework for privacy-preserving training, as it providesformal privacy guarantees. However, compared to their non-private counterparts,models trained with DP often have significantly reduced accuracy. Privateclassifiers are also believed to exhibit larger performance disparities acrosssubpopulations, raising fairness concerns. The poor performance of classifierstrained with DP has prevented the widespread adoption of privacy preservingmachine learning in industry. Here we show that pre-trained foundation modelsfine-tuned with DP can achieve similar accuracy to non-private classifiers,even in the presence of significant distribution shifts between pre-trainingdata and downstream tasks. We achieve private accuracies within a few percentof the non-private state of the art across four datasets, including two medicalimaging benchmarks. Furthermore, our private medical classifiers do not exhibitlarger performance disparities across demographic groups than non-privatemodels. This milestone to make DP training a practical and reliable technologyhas the potential to widely enable machine learning practitioners to trainsafely on sensitive datasets while protecting individuals' privacy.

Unlocking Accuracy and Fairness in Differentially Private Image Classification