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

Exploiting gradient leakage to reconstruct supposedly private training data,gradient inversion attacks are an ubiquitous threat in collaborative learningof neural networks. To prevent gradient leakage without suffering from severeloss in model performance, recent work proposed a PRivacy EnhanCing mODulE(PRECODE) based on variational modeling as extension for arbitrary modelarchitectures. In this work, we investigate the effect of PRECODE on gradientinversion attacks to reveal its underlying working principle. We show thatvariational modeling induces stochasticity on PRECODE's and its subsequentlayers' gradients that prevents gradient attacks from convergence. Bypurposefully omitting those stochastic gradients during attack optimization, weformulate an attack that can disable PRECODE's privacy preserving effects. Toensure privacy preservation against such targeted attacks, we propose PRECODEwith Partial Perturbation (PPP), as strategic combination of variationalmodeling and partial gradient perturbation. We conduct an extensive empiricalstudy on four seminal model architectures and two image classificationdatasets. We find all architectures to be prone to gradient leakage, which canbe prevented by PPP. In result, we show that our approach requires lessgradient perturbation to effectively preserve privacy without harming modelperformance.

Combining Variational Modeling with Partial Gradient Perturbation to Prevent Deep Gradient Leakage