Back to feed
arXiv cs.LG
arXiv cs.LG
7/9/2026
WHERE to Generate Matters: Budget-Aware Synthetic Augmentation for Label Skewed Federated Learning

WHERE to Generate Matters: Budget-Aware Synthetic Augmentation for Label Skewed Federated Learning

Short summary

Label skew in federated learning causes client drift and reduces global model accuracy. The authors propose FedEAS, an entropy-adaptive policy that assigns per-client, per-class synthetic data generation budgets based on local label distributions, jointly deciding how much to generate and where to allocate samples. FedEAS recovers most of the accuracy gains of full class balancing while cutting generation cost by 94.1% and outperforming uniform allocation by up to 18.82% on CIFAR-10/100.

  • •FedEAS assigns entropy-adaptive per-class generation budgets to each FL client based on local label distribution
  • •Reduces generation budget by 94.1% while retaining most accuracy gains of full class balancing
  • •Outperforms uniform allocation by up to 18.82% at equal total budget on CIFAR-10 and CIFAR-100

Generated with AI, which can make mistakes.

#research-breakthrough
Read full article at arXiv cs.LG

Is this a good recommendation for you?

Comments

Failed to load comments. Please try again.

Explore more

Federated Explainable Artificial Intelligence: Roles, Architectures, Evaluation, and Open Challenges
arXiv cs.LG
Post-Generation Curation of Synthetic Images via Homogeneous-Heterogeneous Splitting
arXiv cs.LG
Federated Learning for Object Detection: Enabling Collaborative Drone Learning Without Centralizing Data
arXiv cs.LG
HantaWatch: Federated Learning for Hantavirus Genomic Surveillance
arXiv cs.LG