Explainable AI Approaches in Federated Learning: Systematic Review.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Data Science

Background:

• Generative AI systems like ChatGPT and Bard have increased AI accessibility.
• Understanding machine learning (ML) model operations is crucial for trust.
• Explainable AI (XAI) addresses the challenge of understanding AI model outputs.
• Federated Learning (FL) enables privacy-preserving AI through decentralized models.

Purpose of the Study:

• To assess the development of explainable AI within the federated learning environment.
• To identify key contributions, FL types, application sectors, models, and methods in explainable FL research.
• To catalog the databases used for sourcing studies in this field.

Main Methods:

• A systematic literature search was conducted across eight major electronic databases.
• Databases included Web of Science, Scopus, PubMed, ACM Digital Library, IEEE Xplore, Mendeley, BASE, and Google Scholar.

Main Results:

• Research on explainable FL is expanding, with a concentration in Europe and Asia.
• Data privacy and limited training data are primary drivers for FL adoption.
• Horizontal FL is the predominant FL approach.
• Post hoc explainability techniques are favored in current research.

Conclusions:

• The field of explainable FL shows significant potential for developing new and enhancing existing methods.
• Improvements are particularly needed for critical application areas.
• Further research can foster greater trust and adoption of AI in sensitive domains.