XAI for debugging ML models

This study discusses the use of explainable AI (XAI) methods for debugging AI models. While XAI methods ideally simplify the debugging of AI models, the selection of the best XAI method for a given use case is still a problem at the moment. To help with this, this paper provides an overview of categorization approaches for XAI methods, addresses the evaluation problems of current explanation methods and presents the current state of research on XAI for debugging based on seven recent studies. Finally, general recommendations for the use of XAI for debugging are formulated.