FedReLa: Imbalanced Federated Learning via Re-Labeling

Federated learning has emerged as the foremost approach for decentralized model training with privacy preserving. 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.

AI training across separate devices protects user privacy, but it often struggles with uneven data categories and inconsistent data across devices. When some devices lack certain data types entirely, combined issues make the shared AI model much less accurate. To fix this, we created a simple method called FedReLa. It adjusts data labels based on inherent features to fix the model’s biased judgments, without needing to know the overall data distribution. It works with existing tools and adds no extra data transfer burden. Our tests prove FedReLa outperforms current methods. It greatly boosts the model’s minority classes and overall performance and its ability to recognize rare data categories on imbalanced and diversely distributed datasets.