AI Paper Atlas

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

FedBE: Making Bayesian Model Ensemble Applicable to Federated Learning

Hong-You Chen, Wei-Lun Chao

ICLR 2021 · Federated learning and privacy › Personalized and heterogeneous FL · approach: Non-IID federated learning · 28 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
Replace weight averaging with Bayesian model sampling, ensemble, and server-side distillationFederated learning · CIFAR-10, non-i.i.d. Step, ResNet20test accuracy (%): 70.2$\pm$0.17 -> 77.5$\pm$0.42Improved
Add SWA to ensemble distillation (ablation)FedBE · CIFAR-10, non-i.i.d. Step, ResNet20test accuracy (%): 74.9$\pm$1.41 -> 77.5$\pm$0.42Improved
Add sampled global models to ensemble distillation targets (ablation)FedBE · CIFAR-10, non-i.i.d. Step, ConvNettest accuracy (%): 72.6$\pm$0.28 -> 73.1$\pm$0.46Improved
Combine sampled models, client weight average, and client models as distillation targets (ablation)FedBE · CIFAR-10, non-i.i.d. Step, ConvNettest accuracy (%): 74.0$\pm$0.66 -> 74.5$\pm$0.51Improved
Replace SWA with SGD for Bayesian ensemble distillation (ablation)FedBE · CIFAR-10, non-i.i.d. Step, ConvNet, one-roundtest accuracy (%): 35.7$\pm$0.86 -> 35.1$\pm$0.76Hurt
Sample models from the Bayesian distribution rather than ensemble only client models (ablation)Model ensemble · CIFAR-10, non-i.i.d. Step, ConvNet, one-roundtest accuracy (%): 60.5$\pm$0.28 -> 62.5$\pm$0.35Improved
Use out-of-domain CIFAR-100 or Tiny-ImageNet unlabeled data instead of CIFAR-10 server data (ablation)FedBE · CIFAR-10, non-i.i.d. Steptest accuracy (%): 74.5$\pm$0.51; 74.4$\pm$0.45; 74.5$\pm$0.64No clear effect
Increase the amount of unlabeled server data (ablation)FedBE · CIFAR-10, non-i.i.d. Steptest accuracy (%)Improved
Add FedBE aggregation to FedProxFedProx · CIFAR-10, non-i.i.d. Step, ResNet20test accuracy (%): 71.1$\pm$0.52 -> 77.7$\pm$0.45Improved
Add FedBE aggregation to FedAvgMFedAvgM · CIFAR-10, non-i.i.d. Step, ResNet20test accuracy (%): 73.2$\pm$0.57 -> 78.0$\pm$0.46Improved
Replace batch normalization with group normalization (ablation)FedAvg ResNet20 · CIFAR-10, non-i.i.d. Step, 200 roundstest accuracy (%): 74.6% -> 76.4%Improved
Apply FedBE aggregation to SCAFFOLDSCAFFOLD · CIFAR-10, non-i.i.d. Step, ResNet20test accuracy (%): 59.4% -> 76.4%Improved

Later papers that compared with this method

3 later accepted papers report FedBE as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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