A Survey of Recent Advances for Tackling Data Heterogeneity in Federated Learning

Federated Learning (FL), as a distributed machine learning framework, allows multiple parties to collaboratively train models without sharing their data, thereby protecting privacy and data security. However, the issue of data heterogeneity—where data distributions, feature spaces, and label spaces vary significantly across different clients—poses a critical challenge to the effectiveness of federated learning. To address this problem, researchers have proposed various solutions, including techniques to mitigate local model drift, adaptive model aggregation, local data augmentation, and personalized federated learning. These strategies collectively enhance the capability of federated learning in handling data heterogeneity, promoting its widespread application across numerous fields. This review aims to summarize and discuss the latest advancements in these technologies.