Answer-Set Programs for Reasoning about Counterfactual Interventions and Responsibility Scores for Classification

TL;DR

This paper introduces a method using answer-set programming to model counterfactual interventions and compute responsibility scores, enhancing explainability of classification models with domain knowledge integration.

Contribution

It presents a novel declarative approach using answer-set programs to reason about counterfactuals and responsibility scores in classification, including domain knowledge incorporation.

Findings

Effective modeling of counterfactual interventions

Ability to compute responsibility scores for explanations

Supports domain knowledge and query answering

Abstract

We describe how answer-set programs can be used to declaratively specify counterfactual interventions on entities under classification, and reason about them. In particular, they can be used to define and compute responsibility scores as attribution-based explanations for outcomes from classification models. The approach allows for the inclusion of domain knowledge and supports query answering. A detailed example with a naive-Bayes classifier is presented.