Exploring and Evaluating Classifier Decisions with Concept- and Relation-based Explanations (CoReX)
Combining concept extraction from CNN feature maps with interpretable relational learning can provide faithful and insightful explanations for CNN image classification decisions.
