Owen-based Semantics and Hierarchy-Aware Explanation (O-Shap)
This work addresses the challenge of interpretability in explainable AI for vision applications, offering an incremental improvement over existing methods by enhancing semantic alignment in hierarchical feature attributions.
The paper tackles the problem of feature dependencies in Shapley value-based explanations for vision tasks by proposing a new segmentation approach that satisfies the T-property for semantic alignment, resulting in improved attribution precision, semantic coherence, and runtime efficiency compared to baseline SHAP variants.
Shapley value-based methods have become foundational in explainable artificial intelligence (XAI), offering theoretically grounded feature attributions through cooperative game theory. However, in practice, particularly in vision tasks, the assumption of feature independence breaks down, as features (i.e., pixels) often exhibit strong spatial and semantic dependencies. To address this, modern SHAP implementations now include the Owen value, a hierarchical generalization of the Shapley value that supports group attributions. While the Owen value preserves the foundations of Shapley values, its effectiveness critically depends on how feature groups are defined. We show that commonly used segmentations (e.g., axis-aligned or SLIC) violate key consistency properties, and propose a new segmentation approach that satisfies the $T$-property to ensure semantic alignment across hierarchy levels. This hierarchy enables computational pruning while improving attribution accuracy and interpretability. Experiments on image and tabular datasets demonstrate that O-Shap outperforms baseline SHAP variants in attribution precision, semantic coherence, and runtime efficiency, especially when structure matters.