A Comprehensive Review of Federated Learning: Privacy, Multimodality, and Robustness in Heterogeneous Environments

Abstract

Federated Learning (FL) enables collaborative model training while preserving data privacy by keeping raw data locally on client devices. Despite its advantages, practical FL deployments face major challenges due to statistical heterogeneity, multimodal data distributions, missing modalities, and the trade–off between privacy, utility and efficiency. Existing methods typically address these challenges in isolation, resulting in performance degradation, excessive overhead, or weakened privacy guarantees. This paper presents a comprehensive review of recent advances in federated learning through a pillar-based analysis. A technical taxonomy is introduced to categorize aggregation strategies, privacy mechanisms, and modality handling approaches. Key research gaps are identified, highlighting the absence of unified frameworks that simultaneously address heterogeneity, multimodality, and strong privacy. Finally, future research directions toward practical and trustworthy federated learning systems are discussed.