AI Paper Atlas

Research map / Federated learning and privacy / Communication-efficient FL

DISTREAL: Distributed Resource-Aware Learning in Heterogeneous Systems

Martin Rapp, Ramin Khalili, Kilian Pfeiffer, Jörg Henkel

AAAI 2022 · Federated learning and privacy › Communication-efficient FL · approach: Decentralized peer-to-peer training · 12 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Use Pareto-optimized, per-layer filter dropout instead of one shared dropout rateDISTREAL · CIFAR-10 and CIFAR-100, heterogeneous resources across devicesconvergence speedImproved
Adjust the dropout vector locally before each mini-batch as resources changeDISTREAL · CIFAR-10 and CIFAR-100, heterogeneous resources across devices and timeconvergence speedImproved
Use a smaller network sized for the weakest deviceFederated learning · CIFAR-10, heterogeneous resources across devicesearly convergence speedImproved
Use a smaller network sized for the weakest deviceFederated learning · FEMNIST, heterogeneous resources across devicesconvergence speedHurt
Change from a shared dropout rate to optimized per-layer dropout rates (ablation)DISTREAL · DenseNet-40 design-space explorationtrade-off between MAC requirements and convergence speedImproved
Change from fixed dropout throughout a round to resource-aware selection at each mini-batch (ablation)DISTREAL · CIFAR-10 and CIFAR-100, time-varying resourcesconvergence speedImproved
Increase the rate of resource changes during a roundFederated Dropout and HeteroFL · CIFAR-10 and CIFAR-100, heterogeneous resources across devices and timeconvergence speedHurt
Increase the rate of resource changes during a roundDISTREAL · CIFAR-10 and CIFAR-100, heterogeneous resources across devices and timeconvergence speedNo clear effect
Apply filter-based structured dropout to convolutional layersNeural network training · DenseNet-40 on Raspberry Pi 4training timeImproved
Use a per-layer dropout vector with rates limited to [0,0.5] (ablation)DISTREAL · Design-time searchfinal accuracyImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy