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 (PPML) aims at enabling machine learning(ML) algorithms to be used on sensitive data. We contribute to this line ofresearch by proposing a framework that allows efficient and secure evaluationof full-fledged state-of-the-art ML algorithms via secure multi-partycomputation (MPC). This is in contrast to most prior works, which substitute MLalgorithms with approximated "MPC-friendly" variants. A drawback of the latterapproach is that fine-tuning of the combined ML and MPC algorithms is required,which might lead to less efficient algorithms or inferior quality ML. This isan issue for secure deep neural networks (DNN) training in particular, as thisinvolves arithmetic algorithms thought to be "MPC-unfriendly", namely, integerdivision, exponentiation, inversion, and square root. In this work, we proposesecure and efficient protocols for the above seemingly MPC-unfriendlycomputations. Our protocols are three-party protocols in the honest-majoritysetting, and we propose both passively secure and actively secure with abortvariants. A notable feature of our protocols is that they simultaneouslyprovide high accuracy and efficiency. This framework enables us to efficientlyand securely compute modern ML algorithms such as Adam and the softmax function"as is", without resorting to approximations. As a result, we obtain secure DNNtraining that outperforms state-of-the-art three-party systems; our fulltraining is up to 6.7 times faster than just the online phase of the recentlyproposed FALCON@PETS'21 on a standard benchmark network. We further performmeasurements on real-world DNNs, AlexNet and VGG16. The performance of ourframework is up to a factor of about 12-14 faster for AlexNet and 46-48 fasterfor VGG16 to achieve an accuracy of 70% and 75%, respectively, when compared toFALCON.

Adam in Private: Secure and Fast Training of Deep Neural Networks with Adaptive Moment Estimation