AI Paper Atlas

Research map / Federated learning and privacy / Personalized and heterogeneous FL

Fair-FedMOE: Group-Fair One-Shot Federated Learning via Prototype-Guided Experts for Medical Imaging Analysis

Lingzhao Meng, Shuai Guo, Weishan Zhang, Zengxiang Li, Han Yu, Nan Liu, Daniel Shu Wei Ting, Yuru Liu, Shudong Wang, Tao Chen

ICML 2026 · Federated learning and privacy › Personalized and heterogeneous FL · approach: Federated medical and multimodal learning

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 Fairness-aware Expert Routing and Prototype-guided Differential Aggregation (full Fair-FedMOE) vs best one-shot baselineFair-FedMOE · DR detection, Age x GenderRES-AUC: +8.15 vs best OFL (RETFound)Improved
add prototype-guided differential aggregation (PDA) (ablation)Fair-FedMOE · DR detection, Age x Gender, RETFoundRES-AUC: 77.92 -> 82.32Improved
add fairness-aware expert routing (FER) (ablation)Fair-FedMOE · DR detection, Age x Gender, RETFoundRES-AUC: 80.89 -> 82.32Improved
add FER to local training without aggregation (ablation)Local training · DR detection, Age x Gender, RETFoundAUC: 86.75 -> 87.29Improved
add prototype constraint loss Lpc (ablation)Fair-FedMOE · DR detection, Age x Gender, RETFoundRES-AUC: 78.49 -> 82.32Improved
add sign consistency loss Lsign (ablation)Fair-FedMOE · DR detection, Age x Gender, RETFoundRES-AUC: 79.49 -> 82.32Improved
add l1 sparsity regularization on experts (ablation)Fair-FedMOE · DR detection, Age x Gender, RETFoundRES-AUC: 79.59 -> 82.32Improved
use Fair-FedMOE one-shot instead of local trainingFair-FedMOE · Glaucoma detection, Age x Gender, RETFoundRES-AUC: 72.60 -> 76.54Improved
use FedAvg (traditional FL) instead of local trainingFedAvg · DR detection, Age x Gender, RETFoundRES-AUC: 79.28 -> 71.43Hurt
increase number of groups from 6 to 18 (add race)FlexFair vs Fair-FedMOE · Glaucoma detection, RETFoundRES-AUC: FlexFair 75.14 (gender) vs 64.07 (18 groups); Fair-FedMOE keeps gainsMixed
switch fairness metric from ES-AUC to RES-AUC as group count rises from 2 to 18fairness metric · Glaucoma detection (RETFound)metric value: ES-AUC 78.7% -> 38.7%; RES-AUC 79.9% -> 71.0%Improved
set prototype-similarity weight beta to 0.7 in aggregation (ablation)PDA · DR detection, RETFoundRES-AUCMixed

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