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

Federated learning (FL) is a framework for users to jointly train a machinelearning model. FL is promoted as a privacy-enhancing technology (PET) thatprovides data minimization: data never "leaves" personal devices and usersshare only model updates with a server (e.g., a company) coordinating thedistributed training. While prior work showed that in vanilla FL a maliciousserver can extract users' private data from the model updates, in this work wetake it further and demonstrate that a malicious server can reconstruct userdata even in hardened versions of the protocol. More precisely, we propose anattack against FL protected with distributed differential privacy (DDP) andsecure aggregation (SA). Our attack method is based on the introduction ofsybil devices that deviate from the protocol to expose individual users' datafor reconstruction by the server. The underlying root cause for thevulnerability to our attack is a power imbalance: the server orchestrates thewhole protocol and users are given little guarantees about the selection ofother users participating in the protocol. Moving forward, we discussrequirements for privacy guarantees in FL. We conclude that users should onlyparticipate in the protocol when they trust the server or they apply localprimitives such as local DP, shifting power away from the server. Yet, thelatter approaches come at significant overhead in terms of performancedegradation of the trained model, making them less likely to be deployed inpractice.

Reconstructing Individual Data Points in Federated Learning Hardened with Differential Privacy and Secure Aggregation