Aumann-SHAP: The Geometry of Counterfactual Interaction Explanations in Machine Learning
Researchers have developed Aumann-SHAP, a new framework for explaining machine learning model decisions by analyzing counterfactual interactions. This method decomposes changes by focusing on a local hypercube between baseline and counterfactual features, discretizing it into a grid to form a cooperative game. Shapley and LES values applied to this game provide geometry-aware attributions that converge to the Integrated Gradients limit and can be computed efficiently. AI
IMPACT Introduces a novel method for explaining ML model behavior, potentially improving interpretability and trust in AI systems.