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 widespread adoption of handheld devices have fueled rapid growth in newapplications. Several of these new applications employ machine learning modelsto train on user data that is typically private and sensitive. FederatedLearning enables machine learning models to train locally on each handhelddevice while only synchronizing their neuron updates with a server. While thisenables user privacy, technology scaling and software advancements haveresulted in handheld devices with varying performance capabilities. Thisresults in the training time of federated learning tasks to be dictated by afew low-performance straggler devices, essentially becoming a bottleneck to theentire training process. In this work, we aim to mitigate the performancebottleneck of federated learning by dynamically forming sub-models forstragglers based on their performance and accuracy feedback. To this end, weoffer the Invariant Dropout, a dynamic technique that forms a sub-model basedon the neuron update threshold. Invariant Dropout uses neuron updates from thenon-straggler clients to develop a tailored sub-models for each stragglerduring each training iteration. All corresponding weights which have amagnitude less than the threshold are dropped for the iteration. We evaluateInvariant Dropout using five real-world mobile clients. Our evaluations showthat Invariant Dropout obtains a maximum accuracy gain of 1.4% points overstate-of-the-art Ordered Dropout while mitigating performance bottlenecks ofstragglers.

Reducing Impacts of System Heterogeneity in Federated Learning using Weight Update Magnitudes