AI Paper Atlas

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

Class-Grouped Normalized Momentum and Faster Hyperparameter Exploration to Tackle Class Imbalance in Federated Learning

Haemin Park, Diego Klabjan, Martin W. Braun, Xiuqi Li, Balakrishnan Ananthanarayanan

ICML 2026 · Federated learning and privacy › Communication-efficient FL · approach: Client participation analysis · 0 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
add group momentum (beta=0.5) to class-grouped normalization (ablation)FedCGN · CIFAR-10-LT imbalance 20 (K=5)accuracy (%): 83.35 -> 85.65Improved
add group momentum (beta=0.5) to class-grouped normalization (ablation)FedCGN · CIFAR-100-LT imbalance 20 (K=20)accuracy (%): 38.27 -> 40.51Improved
raise momentum factor from 0.5 to 0.9 (ablation)FedCGNM · CIFAR-10-LT imbalance 20 (K=5)accuracy (%): 85.65 -> 83.93Hurt
use within-group variance grouping instead of naive half-split (ablation)FedCGNM · CIFAR-10-LT imbalance 20 (K=5)F1: 0.8382 -> 0.8565Improved
use within-group variance grouping instead of naive half-split (ablation)FedCGNM · CIFAR-100-LT imbalance 20 (K=5)F1: 0.4339 -> 0.4983Improved
use 3 or 4 groups instead of 2 (ablation)FedCGNM · CIFAR-100-LTvalidation loss / accuracyNo clear effect
add FedHOO sampling-rate search to uniform samplingFedCGNM · CIFAR-10-LT imbalance 100 (K=5, IID)F1: 0.7432 -> 0.7485Improved
add FedHOO sampling-rate search to uniform samplingFedCGNM · CIFAR-10-LT imbalance 100 (K=5, Non-IID)F1: 0.5896 -> 0.5945Improved
replace per-class normalization (FedCGN) by grouped normalized momentum (FedCGNM)FedCGN · CIFAR-10-LT imbalance 100 (K=5, IID)F1: 0.7054 -> 0.7432Improved
use class-weighted cross-entropy instead of plain FedAvgFedAvg · CIFAR-10-LT imbalance 100 (K=5, IID)F1: 0.6177 -> 0.5379Hurt
increase clients from K=5 to K=20FedCGNM + U · CIFAR-10-LT imbalance 100 (IID)F1: 0.7432 -> 0.5294Hurt
use loss reweighting (Ratio Loss) under mild skewRatio Loss · Adult Income (mild imbalance)F1No 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