Declarative Approaches to Counterfactual Explanations for Classification

Abstract We propose answer-set programs that specify and compute counterfactual interventions on entities are input a classification model. In relation to the outcome of model, resulting serve as basis for definition computation causality-based explanation scores feature values in entity under classification, namely responsibility . The approach can be applied with black-box models, also models specified logic programs, such rule-based classifiers. main focus this study is specification best entities, is, those lead maximum scores. From them one read off explanations original entity. extend bring into picture semantic or domain knowledge. show how could extended by means probabilistic methods, underlying probability distributions modified through use constraints. Several examples written syntax DLV ASP-solver, run it, shown.