AI Paper Atlas

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

FedAdamom: Adaptive Momentum for Improved Generalization in Federated Optimization

Wenjie Hou, Tianxiang Chen, Feng Wang, Tiantong Wu, Zhiming Zheng, Shaoting Tang, Wei Yang Bryan Lim

CVPR 2026 · Federated learning and privacy › Communication-efficient FL · approach: Local drift correction

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
adapt the momentum parameter instead of the learning rate in the global server updateFedAdam · CIFAR-10 (ResNet-18, 100 clients)accuracy after 1000 rounds (%): 87.31 -> 88.93Improved
adaptive momentum instead of adaptive learning rateFedAdam · CIFAR-100 (ResNet-18, 100 clients)accuracy after 1000 rounds (%): 53.67 -> 57.58Improved
adaptive momentum instead of adaptive learning rateFedAdam · Tiny-ImageNet (ResNet-18, 100 clients)accuracy after 1000 rounds (%): 41.75 -> 47.38Improved
adaptive momentum instead of adaptive learning rateFedAdam · CIFAR-100 (ResNet-18)rounds to 45% accuracy: 390 -> 276Improved
adaptive momentum instead of adaptive learning rateFedAdam · CIFAR-10 global loss landscape (CNN, Dir(0.3))flatness of global minimumImproved
add server-side global momentumFedAvg · escape from sharp minima (theory)mean escape timeNo clear effect
add global momentum to FedAvgFedAvg · CIFAR-100 (ResNet-18)accuracy after 500 rounds (%): 46.74 -> 50.87Improved
increase beta2 from 0.05 to 0.3 (ablation)FedAdamom · CIFAR-10 Dir(0.3)accuracy (%): 88.93 -> 88.13No clear effect
increase beta2 from 0.05 to 0.3 (ablation)FedAdamom · CIFAR-10 i.i.d.accuracy (%): 91.83 -> 91.32No clear effect
use client-side adaptive momentum (FedCAda) instead of server-sideFedAvg baselines · CIFAR-100 (ResNet-18)accuracy after 1000 rounds (%): 54.10 -> 47.51Hurt

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