A Geometric View of Counterfactual Behavior: Interaction of Boundary Proximity and Local Support
Researchers have developed a new method to understand how machine learning models generate counterfactual explanations. Their geometric approach analyzes the interaction between decision boundary proximity and local data support within model representations. This analysis reveals that models with similar predictive accuracy can exhibit significantly different counterfactual behaviors, suggesting that counterfactual properties are a distinct dimension beyond mere performance. AI
IMPACT Provides a new framework for understanding and potentially improving the interpretability and robustness of machine learning models.