Hierarchical Concept-based Interpretable Models

Researchers introduce Hierarchical Concept Embedding Models (HiCEMs), a new approach to make deep neural networks more interpretable by modeling relationships between concepts in hierarchical structures. The method includes Concept Splitting to automatically discover fine-grained sub-concepts without additional annotations, reducing the burden of manual labeling while improving model accuracy and interpretability.

• →HiCEMs address limitations of existing Concept Embedding Models by explicitly modeling inter-concept relationships through hierarchical structures.
• →Concept Splitting method automatically discovers interpretable sub-concepts from pretrained models without requiring additional manual annotations.
• →The approach enables fine-grained explanations from limited concept labels, reducing annotation costs for training interpretable models.
• →Testing on multiple datasets including a new PseudoKitchens dataset shows HiCEMs can discover human-interpretable concepts absent during training.
• →HiCEMs support test-time concept interventions at different granularities, leading to improved task accuracy.