Towards a Game-Theoretic View of Baseline Values in the Shapley Value

Jie Ren, Zhanpeng Zhou, Qirui Chen, Quanshi Zhang

Abstract: This paper aims to formulate the problem of estimating optimal baseline values, which are used to compute the Shapley value in game theory. In the computation of Shapley values, people usually set an input variable to its baseline value to represent the absence of this variable. However, there are no studies on how to ensure that baseline values represent the absence states of variables without bringing in additional information, which ensures the trustworthiness of the Shapley value. To this end, previous studies usually determine baseline values in an empirical manner, which are not reliable. Therefore, we revisit the feature representation of a deep model in game theory, and formulate the absence state of an input variable. From the perspective of game-theoretic interaction, we learn the optimal baseline value of each input variable. Experimental results have demonstrated the effectiveness of our method. The code will be released when the paper is accepted.