Towards Attributions of Input Variables in a Coalition

TL;DR

This paper investigates how to meaningfully partition input variables for attribution in Explainable AI, extending Shapley values to better handle variable coalitions and interactions, validated across diverse AI tasks.

Contribution

It introduces a new attribution metric for variable coalitions, analyzes interaction effects, and proposes metrics for coalition faithfulness, addressing theoretical gaps in variable partitioning.

Findings

Identifies attribution conflicts caused by variable interactions.

Proposes three metrics to evaluate coalition faithfulness.

Validates approach across multiple AI domains with consistent results.

Abstract

This paper focuses on the fundamental challenge of partitioning input variables in attribution methods for Explainable AI, particularly in Shapley value-based approaches. Previous methods always compute attributions given a predefined partition but lack theoretical guidance on how to form meaningful variable partitions. We identify that attribution conflicts arise when the attribution of a coalition differs from the sum of its individual variables' attributions. To address this, we analyze the numerical effects of AND-OR interactions in AI models and extend the Shapley value to a new attribution metric for variable coalitions. Our theoretical findings reveal that specific interactions cause attribution conflicts, and we propose three metrics to evaluate coalition faithfulness. Experiments on synthetic data, NLP, image classification, and the game of Go validate our approach, demonstrating consistency with human intuition and practical applicability.