AI Paper Atlas

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

ShapCCS: Shapley-Driven Client Coreset Selection in Federated Learning

Shuo Ji, Jie Hu, Zhouqiao He, Zijie Zhao, Tianrui Li, Jie Xu

ICML 2026 · Federated learning and privacy › Communication-efficient FL · approach: Communication-efficient heterogeneous 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 ShapCCS client-level coreset to Random data selection (ablation)Random · CIFAR-10 Dir(0.1), r=70%accuracy: +9.24 (62.22)Improved
add ShapCCS client-level coreset to Random data selection (ablation)Random · CIFAR-10 Dir(0.1), r=30%accuracy: +15.22 (55.66)Improved
add ShapCCS to FedCS (ablation)FedCS · CIFAR-10 Dir(0.1), r=70%accuracy: +10.14 (60.77)Improved
add ShapCCS to existing coreset baselines (ablation)CRAIG, EL2N, CCS, Moderate, GCFL, FedCS · CIFAR-10 Dir(0.1)accuracy: gains of 2.67% to 15.22%Improved
use ShapCCS instead of ModerateModerate · CIFAR-10 IID, r=30%accuracy: 62.75 -> 72.59Improved
use ShapCCS instead of ModerateModerate · CIFAR-100 Dir(0.1), r=30%accuracy: 28.15 -> 35.79Improved
use ShapCCS with accuracy surpassing full datasetShapCCS · CIFAR-10 Dir(0.1), r=70%accuracy: 61.51 -> 62.72Improved
drop data-level selection (ShapCCS* vs ShapCCS) (ablation)ShapCCS · CIFAR-10 Dir(0.5), r=30%accuracy: 65.16 -> 65.61No clear effect
use ShapCCS* with few clients and small client number (ablation)ShapCCS* · CIFAR-10 Dir(0.1), r=30%, small |N|accuracyHurt
retain all clients (small threshold tau) instead of excluding fragmented clients (ablation)ShapCCS · CIFAR-10 Dir(0.1), r=30%accuracyHurt
tune threshold tau up to B/20 (ablation)ShapCCS · CIFAR-10 Dir(0.1), r=30%accuracyImproved
replace GPSV with GTG-SV, NSV, CGSV or CSSV Shapley variants (ablation)ShapCCS* · CIFAR-10 Dir(0.1)accuracyNo 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