AI Paper Atlas

Research map / Federated learning and privacy / Personalized and heterogeneous FL

FjORD: Fair and Accurate Federated Learning under heterogeneous targets with Ordered Dropout

Samuel Horváth, Stefanos Laskaridis, Mario Almeida, Ilias Leontiadis, Stylianos Venieris, Nicholas Donald Lane

NeurIPS 2021 · Federated learning and privacy › Personalized and heterogeneous FL · approach: Federated learning with pretrained models · 64 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 ordered nested submodels instead of random dropoutFjORD · CIFAR10accuracy: 1.53-34.87 percentage points (pp) ($19.22$ pp avg. across $p$ values)Improved
use ordered nested submodels instead of random dropoutFjORD · FEMNISTaccuracy: 1.57-6.27 pp ($3.41$ pp avg.)Improved
use ordered nested submodels instead of random dropoutFjORD · Shakespearenegative perplexity: 0.01-0.82 points (p) ($0.46$ p avg.)Improved
add eFD random dropout (ablation)FjORD · all three datasetsaccuracy / negative perplexity: 0.71-2.66 pp ($1.79$ avg.), up to $2.56$ pp ($1.35$ pp avg.), 0.12-$0.22$ p ($0.18$ p avg.)Hurt
add knowledge distillation (ablation)FjORD · CIFAR10, p > 0.4accuracy: $0.18, 0.68$ and $0.87$ pp for submodels of size $0.6, 0.8$ and $1$Improved
add knowledge distillation (ablation)FjORD · FEMNISTaccuracy: $1.96, 2.39$ and $2.65$ ppImproved
add knowledge distillation (ablation)FjORD · Shakespearenegative perplexity: $0.10$ pImproved
add knowledge distillation (ablation)FjORD · CIFAR10 and FEMNIST, p <= 0.4accuracyNo clear effect
increase p granularity from uniform-5 to uniform-10 (ablation)FjORD · FEMNIST and Shakespeareaccuracy per submodelNo clear effect
allocate a larger share of clients to the highest-capability cluster (ablation)FjORD · uniform-5, drop scale 0.5 vs 1.0accuracy of smaller submodelsNo clear effect
adapt random dropout rates to client capabilityeFD vs FD · CIFAR10accuracy: 27.13-33 percentage points (pp) (30.94 pp avg.)Improved
adapt random dropout rates to client capabilityeFD vs FD · FEMNISTaccuracy: 4.59-9.04 pp (7.13 pp avg.)Improved

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