AI Paper Atlas

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

OS-Fed: One Snapshot Is All You Need

Xuwei Qian, Jinghui Zhang, Yuchuan Tan, Wenbo Huang, Zhen Wu, Shen Zhou, LiSha Gao, Ding Ding, Fang Dong

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
use OS-FED single snapshot instead of full FedAvg updatesFedAvg (RegNetX-4.0GF) · ImageNet-10total traffic (MB): 8040 MB -> 57 MB, accuracy 0.65 -> 0.64Improved
use multi-grid end-to-end snapshot instead of optimize-then-downsample (ablation)OS-FED · ImageNet-10 (ResNet-18)accuracy: 48.8 -> 63.5Improved
add normalization to the L2 snapshot-matching loss (ablation)OS-FED · ImageNet-10 (ResNet-18)accuracy: 61.0 -> 63.5Improved
use single-level optimization instead of bi-level optimization (ablation)OS-FED · ImageNet-10 (ResNet-18)accuracy: 28.3 -> 61.0Improved
use L2 distance instead of cosine similarity (ablation)OS-FED · ImageNet-10 (ResNet-18)accuracy: 57.4 -> 61.0Improved
add error compensation (ablation)OS-FED · ImageNet-10accuracyImproved
use multi-grid snapshot instead of single monolithic snapshot (ablation)OS-FED · ImageNet-10snapshot qualityImproved
increase grid size from 2x2 to 4x4 (ablation)OS-FED · CIFAR-100accuracy: 26.9% -> 30.3%Improved
add noise during snapshot optimization for privacy (ablation)OS-FED · ImageNet-1K (60 clients)Top-1 accuracy: 35.1% -> 33.5%, CosSim 0.122 -> 0.05Hurt
OS-FED versus DP-SGD for privacyOS-FED · ImageNet-1K (60 clients)wall-time per client: 10.1s -> 3.8sImproved
OS-FED versus RandTopk, LoRA-FAIR on CIFAR-100LoRA-FAIR · CIFAR-100accuracy: 26.3% -> 30.3%Improved
OS-FED on FMNIST with MLP versus FedAFFedAF · FMNIST (MLP, 10 clients)accuracy: 84.1% -> 83.9%No clear effect

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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