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

The idea of federated learning is to collaboratively train a neural networkon a server. Each user receives the current weights of the network and in turnssends parameter updates (gradients) based on local data. This protocol has beendesigned not only to train neural networks data-efficiently, but also toprovide privacy benefits for users, as their input data remains on device andonly parameter gradients are shared. But how secure is sharing parametergradients? Previous attacks have provided a false sense of security, bysucceeding only in contrived settings - even for a single image. However, byexploiting a magnitude-invariant loss along with optimization strategies basedon adversarial attacks, we show that is is actually possible to faithfullyreconstruct images at high resolution from the knowledge of their parametergradients, and demonstrate that such a break of privacy is possible even fortrained deep networks. We analyze the effects of architecture as well asparameters on the difficulty of reconstructing an input image and prove thatany input to a fully connected layer can be reconstructed analyticallyindependent of the remaining architecture. Finally we discuss settingsencountered in practice and show that even averaging gradients over severaliterations or several images does not protect the user's privacy in federatedlearning applications in computer vision.

Inverting Gradients -- How easy is it to break privacy in federated learning?