arXiv:2603.14014v2 Announce Type: replace Abstract: We introduce Aumann-SHAP, an interaction-aware framework that decomposes counterfactual transitions by restricting the model to a local hypercube connecting baseline and counterfactual features. Each hypercube is discretized into a grid to construct an induced micro-player cooperative game in which elementary grid-step moves become players. Shapley and LES values on this TU-micro-game yield geometry-aware within-pot attributions that converge to the diagonal Aumann--Shapley / Integrated Gradients limit under grid refinement, and recover equal
The increasing complexity and opacity of advanced AI models are driving a demand for robust and interpretable explanations of their decisions.
Improved interpretability directly impacts AI safety, fairness, and trust, crucial for widespread adoption and regulatory compliance in critical applications.
The introduction of a new, geometry-aware method for explaining counterfactual interactions in AI models provides a more granular and theoretically grounded approach to understanding model behavior.
- · AI ethicists
- · Regulators
- · Organizations deploying critical AI systems
- · AI developers
- · Companies with opaque proprietary AI models
- · AI systems lacking interpretability
Wider adoption of explainable AI (XAI) techniques, especially in high-stakes environments like healthcare and finance.
Increased trust in AI systems leading to faster integration into various industries and potentially new regulatory frameworks.
Standardization of explanation methodologies and benchmarks, driving innovation in AI interpretability research.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.LG