AI Paper Atlas

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

SMoFi: Step-wise Momentum Fusion for Split Federated Learning on Heterogeneous Data

Mingkun Yang, Ran Zhu, Qing Wang, Jie Yang

AAAI 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 SMoFi step-wise momentum fusionFedAvg · CIFAR-10 (Dir 0.2)top-1 accuracy: 77.16 -> 81.82Improved
add SMoFiFedAvg · CIFAR-100 (Dir 0.2)top-1 accuracy: 48.10 -> 53.83Improved
add SMoFiFedAvg · Tiny-ImageNet (Dir 0.2)top-1 accuracy: 33.43 -> 39.73Improved
add SMoFiFedProx · Tiny-ImageNet (Dir 0.2)top-1 accuracy: 34.86 -> 40.79Improved
add SMoFi instead of FedAvgM/SlowMo/FedNAGFedAvg · CIFAR-10 (Dir 0.2)top-1 accuracy: 79.19 (FedAvgM) -> 81.82Improved
add FedNAG local momentumFedAvg · Tiny-ImageNet (Dir 0.2)top-1 accuracy: 33.43 -> 30.94Hurt
add FedAvgMFedAvg · Tiny-ImageNet (Dir 0.2)top-1 accuracy: 33.43 -> 33.58No clear effect
add SlowMoFedAvg · CIFAR-10 (Dir 0.2)top-1 accuracy: 77.16 -> 76.54Hurt
add SMoFiFedAvg · Tiny-ImageNet (Dir 0.2)rounds-to-target speedup: 1.00x -> 10.06xImproved
aggregate SFLV1 server models each step instead of each epoch (ablation)SFLV1 · CIFAR-10 (Dir 0.2)top-1 accuracy: 77.84 -> 68.10Hurt
use smaller staleness factor alpha (e.g. -1) (ablation)SMoFi · CIFAR-10accuracy / convergenceMixed
move cut layer deeper (ablation)SMoFi · CIFAR-10accuracy gainMixed

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