AI Paper Atlas

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

DecFus: Decentralized Layer-wise Fusion with Dynamic Exploration and Exploitation

Li Yang, Jialong Sun, Chuhai Cai, Xinyang Liu, Yichen Li, Bowen Peng, Jialong Li, Bo Liu

ICML 2026 · Federated learning and privacy › Communication-efficient FL · approach: Efficient local training in FL

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 dynamic cutoff (exchange then averaging) instead of reversed cutoff (ablation)DecFus · CIFAR-100 (alpha=0.1), ResNet-50accuracy: 26.28 -> 35.76Improved
add dynamic cutoff instead of fixed high cutoff (ablation)DecFus · SVHN (alpha=0.1), ResNet-50accuracy: 52.68 -> 57.83Improved
add dynamic cutoff instead of fixed low cutoff (ablation)DecFus · CIFAR-100 (alpha=0.1), ResNet-50accuracy: 30.50 -> 35.76Improved
use cosine similarity instead of L2 distance (ablation)DecFus · CIFAR-100 (alpha=0.1), ResNet-50accuracy: 33.05 -> 35.76Improved
use cosine similarity instead of L2 distance (ablation)DecFus · CIFAR-10 (IID), ResNet-50accuracy: 78.75 -> 78.24No clear effect
use cosine similarity instead of KL divergence (ablation)DecFus · SVHN (alpha=0.1), ResNet-50accuracy: 53.58 -> 57.83Improved
use similarity-based grouping instead of random grouping (ablation)DecFus · CIFAR-10 (alpha=0.1), VGG16accuracy: 48.36 -> 54.14Improved
use similarity-based grouping instead of inverse grouping (ablation)DecFus · CIFAR-100 (IID), VGG16accuracy: 58.85 -> 67.52Improved
use DecFus instead of DFLAvgDFLAvg · CIFAR-100 (alpha=0.1), ResNet-50accuracy: 30.55 -> 35.76Improved
use DecFus instead of DFLAvgDFLAvg · SVHN (alpha=0.1), VGG16accuracy: 75.02 -> 72.84Hurt
use EDecFus (communication-efficient) instead of DecFusDecFus · CIFAR-10 (IID), ResNet-50accuracy: 78.24 -> 76.93Hurt
vary hyperparameters around default (ablation)DecFus · SVHN (alpha=1.0), VGG16accuracy: 93.97 -> 91.84No 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