FedEAS Reduces Synthetic Data Generation Costs by 94.1% in Federated Learning
July 9, 2026
FedEAS uses an entropy-adaptive budget to manage label skew in federated learning. This approach recovers most accuracy gains from full class balancing while cutting generation costs by 94.1% and outperforming uniform allocation by up to 18.82% on CIFAR datasets.
HOW THIS AFFECTS YOU
●
builderYou can reduce the computational overhead of synthetic data augmentation in decentralized training pipelines.
●
researcherYou can optimize federated training efficiency by using entropy-adaptive policies rather than fixed generation budgets.