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 was introduced to enable machine learning over largedecentralized datasets while promising privacy by eliminating the need for datasharing. Despite this, prior work has shown that shared gradients often containprivate information and attackers can gain knowledge either through maliciousmodification of the architecture and parameters or by using optimization toapproximate user data from the shared gradients. However, prior datareconstruction attacks have been limited in setting and scale, as most workstarget FedSGD and limit the attack to single-client gradients. Many of theseattacks fail in the more practical setting of FedAVG or if updates areaggregated together using secure aggregation. Data reconstruction becomessignificantly more difficult, resulting in limited attack scale and/ordecreased reconstruction quality. When both FedAVG and secure aggregation areused, there is no current method that is able to attack multiple clientsconcurrently in a federated learning setting. In this work we introduce LOKI,an attack that overcomes previous limitations and also breaks the anonymity ofaggregation as the leaked data is identifiable and directly tied back to theclients they come from. Our design sends clients customized convolutionalparameters, and the weight gradients of data points between clients remainseparate even through aggregation. With FedAVG and aggregation across 100clients, prior work can leak less than 1% of images on MNIST, CIFAR-100, andTiny ImageNet. Using only a single training round, LOKI is able to leak 76-86%of all data samples.

LOKI: Large-scale Data Reconstruction Attack against Federated Learning through Model Manipulation