New FedXAI Survey: 68-Page Framework for Trustworthy Federated Learning

Federated Learning (FL) trains models across distributed data sources without moving raw data, preserving privacy but not solving the opacity of modern ML models. Meanwhile, Explainable AI (XAI) improves transparency and trust, especially in high-stakes domains. Their intersection – Federated Explainable AI (FedXAI) – aims to satisfy both privacy and explainability requirements. This new survey by Gholizade et al. provides a systematic review of FedXAI, highlighting the shift of explainability from a post-hoc tool to an integral part of the FL lifecycle, supporting aggregation, personalization, robustness, coordination, and system-level decisions.

The authors introduce a taxonomy that classifies FedXAI methods by the role of explainability, model and explainer types, explanation scope, integration level, FL settings, and data heterogeneity. They review approaches from model-agnostic explanations to interpretable federated models and explainability-aware aggregation. The paper also examines evaluation practices, noting a lack of standardized benchmarks and metrics for explanation quality, stability, privacy leakage, and computational overhead. Key challenges identified include explainability under non-IID data, explanation-centric security threats, communication-efficient XAI, continual FedXAI, and integration of domain knowledge and regulatory constraints.