Abstract:Shapley values, which were originally designed to assign attributions to individual players in coalition games, have become a commonly used approach in explainable machine learning to provide attributions to input features for black-box machine learning models. A key attraction of Shapley values is that they uniquely satisfy a very natural set of axiomatic properties. However, extending the Shapley value to assigning attributions to interactions rather than individual players, an interaction index, is non-trivial: as the natural set of axioms for the original Shapley values, extended to the context of interactions, no longer specify a unique interaction index. Many proposals thus introduce additional less ''natural'' axioms, while sacrificing the key axiom of efficiency, in order to obtain unique interaction indices. In this work, rather than introduce additional conflicting axioms, we adopt the viewpoint of Shapley values as coefficients of the most faithful linear approximation to the pseudo-Boolean coalition game value function. By extending linear to $\ell$-order polynomial approximations, we can then define the general family of faithful interaction indices. We show that by additionally requiring the faithful interaction indices to satisfy interaction-extensions of the standard individual Shapley axioms (dummy, symmetry, linearity, and efficiency), we obtain a unique Faithful Shapley Interaction index, which we denote Faith-Shap, as a natural generalization of the Shapley value to interactions. We then provide some illustrative contrasts of Faith-Shap with previously proposed interaction indices, and further investigate some of its interesting algebraic properties. We further show the computational efficiency of computing Faith-Shap, together with some additional qualitative insights, via some illustrative experiments.

The extension of owen coalition value

The Owen Value of Stochastic Cooperative Game

A characterization of the Owen value via sign symmetries

Cooperative networks and f-Shapley value

Absolute Shapley Value

Improved Shapley Values Based on Players’ Least Square Contributions and Their Applications in the Collaborative Profit Sharing of the Rural E-commerce

Shapley Value Approximation Based on Complementary Contribution

The Shapley Value of Cooperative Game with Stochastic Payoffs

Values for cooperative games with a prior unions and a communication graph based on combined effects

The Equal Surplus Division Value for Cooperative Games with a Level Structure

Necessary players and values

Cooperative Games with Overlapping Coalitions

A Coopetition Index for Coalitions in Simple Games

A note on the per capita Shapley support levels value

Value co-creation in open innovation alliances

Coalitional Manipulations and Immunity of the Shapley Value

A Fair-Sharing Value for Cooperative Dynamic Network Games

Cooperative Ressource Sharing With Adamant Player

Optimistic and pessimistic approaches for cooperative games

Faith-Shap: The Faithful Shapley Interaction Index

Approximating the Shapley Value without Marginal Contributions