Survey Reviews Federated Explainable AI Paradigms

Federated Learning (FL) has become a crucial approach for training AI models collaboratively across distributed datasets while preserving data privacy. However, FL alone does not address the inherent opacity of complex machine learning models. Concurrently, Explainable Artificial Intelligence (XAI) has gained prominence for enhancing trust and accountability, especially in critical applications. The convergence of these two fields has given rise to Federated Explainable Artificial Intelligence (FedXAI), which aims to simultaneously meet both privacy and explainability requirements. A comprehensive survey provides a systematic overview of FedXAI, illustrating how explainability is evolving from a post-hoc analysis tool to an integrated component throughout the FL lifecycle. The review introduces a taxonomy to classify FedXAI methods based on various factors, including the role of explainability, model types, explanation scope, and integration levels. It covers diverse approaches, from model-agnostic explanations to inherently interpretable federated models and explainability-aware aggregation mechanisms. The survey also scrutinizes current evaluation practices, noting a significant lack of standardized benchmarks and metrics for assessing explanation quality, stability, privacy leakage, and computational overhead. Finally, it outlines critical open challenges for FedXAI, such as achieving explainability with non-IID data, addressing explanation-centric security threats, developing communication-efficient XAI, and integrating domain knowledge and regulatory compliance. This work serves as a foundational reference for developing transparent, trustworthy, and privacy-preserving federated AI systems.