FedReLa: Imbalanced Federated Learning via Re-Labeling

2026-06-24

AI Summary

This work addresses the performance degradation in federated learning caused by the coexistence of global class imbalance and client data heterogeneity—particularly when certain classes are severely missing on some clients. To tackle this challenge without requiring prior knowledge of the global class distribution, the authors propose a model-agnostic, data-level relabeling approach. The method employs a feature-dependent label reassigner that dynamically adjusts the biased global decision boundary, jointly mitigating data heterogeneity and class imbalance. The proposed framework seamlessly integrates with existing federated learning algorithms without incurring additional communication overhead. Extensive experiments demonstrate significant improvements in both minority-class and overall accuracy on step-imbalanced and long-tailed datasets, outperforming current state-of-the-art methods.

Abstract

Federated learning has emerged as the foremost approach for decentralized model training with privacy preservation. The global class imbalance and cross-client data heterogeneity naturally coexist, and the mismatch between local and global imbalances exacerbates the performance degradation of the aggregated model. The agnosticism of global class distribution poses significant challenges for data-level methods, especially under extreme conditions with severe class absence across clients. In this paper, we propose FedReLa, a novel data-level approach that tackles the coexistence of data heterogeneity and class imbalance in federated learning. By re-labeling samples with a feature-dependent label re-allocator, FedReLa corrects biased global decision boundaries without requiring knowledge of the global class distribution. This modular, model-agnostic approach can be integrated with algorithmic methods to deliver consistent improvements without additional communication overhead. Through extensive experiments, our method significantly improves the accuracy of minority classes and the overall accuracy on stepwise-imbalanced and long-tailed datasets, outperforming the previous state of the art.

Problem

• federated learning
• class imbalance
• data heterogeneity
• global class distribution
• minority classes

Innovation

• Federated Learning
• Class Imbalance
• Data Heterogeneity
• Label Re-allocation
• Model-agnostic