AI Paper Atlas

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

No One Idles: Efficient Heterogeneous Federated Learning with Parallel Edge and Server Computation

Feilong Zhang, Xianming Liu, Shiyi Lin, Gang Wu, Xiong Zhou, Junjun Jiang, Xiangyang Ji

ICML 2023 · Federated learning and privacy › Communication-efficient FL · approach: Communication-efficient heterogeneous FL

Read the paper

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
Swap order so server broadcasts stale global model before aggregation (asynchronous PFL)FedAvg-style FL with average aggregation · CIFAR-10 i.i.dspeedup per global clock: 3.28x vs FedAvg 1xImproved
Synchronous parallel FL (swap aggregation and broadcast order)FedAvg · CIFAR-10 i.i.daccuracy (%): 85.57 vs 85.41No clear effect
Asynchronous parallel FLFedAvg · MNIST / CIFAR-10 / Tiny-ImageNet, i.i.d and non-i.i.daccuracy (%): APFL within about 0.3 of FedAvg on most settingsNo clear effect
Parallel FL with straggler reconnection time ratio Tw/Tl=5FedAvg · CIFAR-10 speedupspeedup: up to 5.56x vs MIFA 3.16xImproved
Parallel FL with distillation-based aggregationFedAvg with distillation aggregation · CIFAR-10, MNIST, Tiny-ImageNet (local epoch 5/10/15)speedup and accuracy: speedups 1.32x to 1.97x; accuracy within about 0.1 of FedAvgImproved
Stale global model in parallel FL (early training)PFL vs FedAvg · CIFAR-10, 20 devicesaccuracy convergence: lower accuracy and stronger oscillations early; similar at convergenceHurt
Asynchronous parallel FL (DGA baseline comparison)Delayed gradient averaging baseline · CIFAR-10 i.i.dspeedup per global clock: 1x for DGANo clear effect
Synchronous parallel FL on MNIST/CIFAR-10/Tiny-ImageNet with stragglers (ablation)FedAvg · Tw/Tl=3 straggler settingspeedup per global clock: SPFL 1.22x-1.33x vs APFL 2.20x-3.91xMixed

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