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

Collaborative training of neural networks leverages distributed data byexchanging gradient information between different clients. Although trainingdata entirely resides with the clients, recent work shows that training datacan be reconstructed from such exchanged gradient information. To enhanceprivacy, gradient perturbation techniques have been proposed. However, theycome at the cost of reduced model performance, increased convergence time, orincreased data demand. In this paper, we introduce PRECODE, a PRivacy EnhanCingmODulE that can be used as generic extension for arbitrary model architectures.We propose a simple yet effective realization of PRECODE using variationalmodeling. The stochastic sampling induced by variational modeling effectivelyprevents privacy leakage from gradients and in turn preserves privacy of dataowners. We evaluate PRECODE using state of the art gradient inversion attackson two different model architectures trained on three datasets. In contrast tocommonly used defense mechanisms, we find that our proposed modificationconsistently reduces the attack success rate to 0% while having almost nonegative impact on model training and final performance. As a result, PRECODEreveals a promising path towards privacy enhancing model extensions.

PRECODE - A Generic Model Extension to Prevent Deep Gradient Leakage