Regression-adjusted Monte Carlo Estimators for Shapley Values and Probabilistic Values

TL;DR

This paper introduces a novel regression-adjusted Monte Carlo method for efficiently estimating probabilistic values like Shapley values, combining sampling and flexible regression techniques to achieve state-of-the-art accuracy in explainable AI.

Contribution

It presents a flexible framework that replaces linear regression with any function family, enabling more accurate and unbiased probabilistic value estimates, including the use of tree-based models.

Findings

State-of-the-art performance on eight datasets

Error reduction up to 6.5x for Shapley values

Error reduction up to 215x for general probabilistic values

Abstract

With origins in game theory, probabilistic values like Shapley values, Banzhaf values, and semi-values have emerged as a central tool in explainable AI. They are used for feature attribution, data attribution, data valuation, and more. Since all of these values require exponential time to compute exactly, research has focused on efficient approximation methods using two techniques: Monte Carlo sampling and linear regression formulations. In this work, we present a new way of combining both of these techniques. Our approach is more flexible than prior algorithms, allowing for linear regression to be replaced with any function family whose probabilistic values can be computed efficiently. This allows us to harness the accuracy of tree-based models like XGBoost, while still producing unbiased estimates. From experiments across eight datasets, we find that our methods give state-of-the-art performance for estimating probabilistic values. For Shapley values, the error of our methods can be $6.5\times$ lower than Permutation SHAP (the most popular Monte Carlo method), $3.8\times$ lower than Kernel SHAP (the most popular linear regression method), and $2.6\times$ lower than Leverage SHAP (the prior state-of-the-art Shapley value estimator). For more general probabilistic values, we can obtain error $215\times$ lower than the best estimator from prior work.