AI Paper Atlas

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

On Large-Cohort Training for Federated Learning

Zachary Charles, Zachary Garrett, Zhouyuan Huo, Sergei Shmulyian, Virginia Smith

NeurIPS 2021 · Federated learning and privacy › Communication-efficient FL · approach: Client participation analysis · 3 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
Increase cohort size from M=10 to M=800FedAvg on EMNIST · EMNISTcatastrophic training failure rate: 0% to 80%Hurt
Apply adaptive clipping to client updatesFedAvg on EMNIST · EMNISTcatastrophic training failure rate: no catastrophic training failures occurred for any cohort sizeImproved
Increase cohort size beyond moderate valuesfederated optimization methods · Multiple taskscommunication-round convergence speedupMixed
Increase cohort size from M=50 to larger cohortsFedAdam · CIFAR-100test accuracyHurt
Increase cohort sizeFedAdam · Stack Overflowaccuracy across test-client percentilesImproved
Scale server learning rate linearly with cohort sizeFedAvg · EMNISTtraining stability: M > 100Hurt
Scale server learning rate by the square root of cohort size, with warmupFedAvg · Multiple taskstrain and test accuracyMixed
Use FedLamb instead of FedAdamfederated training · Large-cohort settingsaccuracyMixed
Use FedLARS instead of coordinate-wise adaptive optimizationfederated training · CIFAR-100, EMNIST, and ShakespeareaccuracyMixed
Start with M=50 and double cohort size every 300 rounds up to M=800FedAvg and FedAdam · Multiple tasksdata efficiencyImproved
Normalize the server pseudo-gradient before the FedAvg server updateFedAvg · Most tasks and cohort sizestraining accuracyImproved
Increase the number of local stepsFedAvg · EMNISTtest accuracyHurt

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