Answer-Set Programs for Reasoning about Counterfactual Interventions and Responsibility Scores for Classification
Artificial Intelligence
2021-09-03 v2 Machine Learning
Logic in Computer Science
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.
Cite
@article{arxiv.2107.10159,
title = {Answer-Set Programs for Reasoning about Counterfactual Interventions and Responsibility Scores for Classification},
author = {Leopoldo Bertossi and Gabriela Reyes},
journal= {arXiv preprint arXiv:2107.10159},
year = {2021}
}
Comments
Revised for camera ready. Extended version with appendices of paper to appear in IJCLR'21. arXiv admin note: text overlap with arXiv:2106.10562