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

2026-07-07

AI Summary

Label skew in federated learning often induces client drift, degrading global model performance. To address this issue, this work proposes FedEAS, a novel method that dynamically links the entropy of each client’s local label distribution to its synthetic data generation budget, thereby determining how much data to generate and to which classes it should be allocated under limited resources. By introducing an entropy-driven adaptive budget allocation mechanism, FedEAS jointly optimizes both the quantity and target classes of synthesized data for efficient and precise data augmentation. Experimental results on CIFAR-10 and CIFAR-100 demonstrate that FedEAS recovers model accuracy close to that achieved under fully balanced data using only 5.9% of the generation budget, and outperforms uniform allocation strategies by up to 18.82% in accuracy under the same budget constraints.

Abstract

Label skew in federated learning (FL) causes client drift and degrades global accuracy. Synthetic data augmentation can reduce this imbalance; however, full class balancing requires substantial computation cost. We propose FedEAS, a policy that assigns each client an entropy-adaptive per-class generation budget computed from its local label distribution. The budget jointly decides \emph{how much} each client generates and \emph{WHERE} the samples go. Accordingly, the total generation budget follows from the per-client budgets rather than being fixed in advance. FedEAS recovers most of the accuracy gain of full class balancing while reducing the generation budget by 94.1\%. At the same total generation budget, it outperforms Uniform allocation by up to 18.82\% across CIFAR-10 and CIFAR-100.

Problem

• label skew
• federated learning
• synthetic data augmentation
• class imbalance
• client drift

Innovation

• federated learning
• label skew
• synthetic data augmentation
• adaptive budget allocation
• entropy-aware generation