AI Paper Atlas

Research map / Optimization / Distributed and compressed

Distributionally Robust Federated Averaging

Yuyang Deng, Mohammad Mahdi Kamani, Mehrdad Mahdavi

NeurIPS 2020 · Optimization › Distributed and compressed · approach: Decentralized optimization algorithms · 18 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 synchronization gap tau from 5 to 15 in DRFA (ablation)DRFA · Fashion MNIST logistic regression, 10 devicescommunication rounds to reach 50% worst-distribution accuracyImproved
increase synchronization gap tau from 5 to 15 in DRFA (ablation)DRFA · Fashion MNIST logistic regression, 10 deviceslocal iterations to reach 50% worst-distribution accuracyHurt
use DRFA (periodic lambda updates) instead of FedAvgfederated logistic regression · Fashion MNIST, worst-distribution accuracyrounds to 50% worst accuracy: FedAvg cannot reach 50% after 300 roundsImproved
use DRFA instead of AFL (tau=1, lambda updated every iteration)AFL · Fashion MNIST, tau=10communication rounds to equal worst accuracyImproved
use DRFA instead of AFLAFL · Fashion MNISTlocal iterations to reach 50% worst accuracyHurt
use DRFA instead of FedAvgFedAvg · Shakespeare, worst-distribution accuracyworst-distribution accuracyNo clear effect
use DRFA instead of FedAvgFedAvg · Fashion MNIST, global training accuracy over 100 roundsaverage training accuracyNo clear effect

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