Causal Explanations for Image Classifiers
Journal of Artificial Intelligence Research · 07.06.2026 · 86
DOI: 10.1613/jair.1.21939 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Artificial Intelligence Research, online erschienen am 07.06.2026. Verfasst von Hana Chockler und 3 weiteren Autor:innen.
Abstract (Originalsprache)
Existing algorithms for explaining the output of image classifiers use different definitions of explanations and a variety of techniques to find them. However, none of the existing tools use a principled approach based on formal definitions of cause and explanation. In this paper we present a novel black-box approach to computing explanations grounded in the theory of actual causality. We prove relevant theoretical results and present an algorithm for computing approximate explanations based on these definitions. We prove termination of our algorithm and discuss its complexity and the amount of approximation compared to the precise definition. We implemented the framework in a tool, ReX, and we present experimental results and a comparison with state-of-the-art tools. We demonstrate that ReX is the most efficient black-box tool and produces the smallest explanations, in addition to outperforming other black-box tools on standard quality measures.
Zitieren
APA
Chockler, H., Kelly, D. A., Kroening, D., & Sun, Y. (2026). Causal Explanations for Image Classifiers. Journal of Artificial Intelligence Research, 86. https://doi.org/10.1613/jair.1.21939
Vancouver
Chockler H, Kelly DA, Kroening D, Sun Y. Causal Explanations for Image Classifiers. Journal of Artificial Intelligence Research. 2026;86. doi:10.1613/jair.1.21939
BibTeX
@article{Chockler2026,
author = {Chockler, Hana and Kelly, David A. and Kroening, Daniel and Sun, Youcheng},
title = {Causal Explanations for Image Classifiers},
journal = {Journal of Artificial Intelligence Research},
year = {2026},
volume = {86},
doi = {10.1613/jair.1.21939}
}
EUVIMED