ExplAIner Language Improves ML Model Explanation Queries

This research introduces ExplAIner, a novel declarative query language designed to enhance the explainability of machine learning classification models. It builds upon existing interpretability frameworks but overcomes significant limitations, particularly in expressing complex optimality-based explanation queries and improving computational efficiency. ExplAIner's layered structure and extended vocabulary allow it to articulate a wide range of explanation types, including abductive, contrastive, feature-based, and distance-based queries. The framework also demonstrates improved evaluation complexity, showing that queries can be processed with a fixed or polynomial number of calls to a SAT solver, making it practical for formal XAI applications.