Rethinking Saliency Maps: A Cognitive Human Aligned Taxonomy and Evaluation Framework for Explanations
Proceedings of the AAAI Conference on Artificial Intelligence · 14.03.2026 · 40 (5) · 3750-3758
DOI: 10.1609/aaai.v40i5.37375 ↗Kurz gefasst
Fachartikel in der Zeitschrift Proceedings of the AAAI Conference on Artificial Intelligence, online erschienen am 14.03.2026. Verfasst von Yehonatan Elisha und 3 weiteren Autor:innen. Die Arbeit wurde bisher 3-mal zitiert (Crossref).
Abstract (Originalsprache)
Saliency maps have become a cornerstone of visual explanation in deep learning, yet there remains no consensus on their intended purpose and their alignment with specific user queries. This fundamental ambiguity undermines both the evaluation and practical utility of explanation methods. In this paper, we introduce the Reference-Frame x Granularity (RFxG) taxonomy—a principled framework that addresses this ambiguity by conceptualizing saliency explanations along two essential axes: the reference-frame axis (distinguishing between pointwise "Why Husky?" and contrastive "Why Husky and not Shih-tzu?" explanations) and the granularity axis (ranging from fine-grained class-level to coarse-grained group-level interpretations, e.g., “Why Husky?” vs. “Why Dog?”). Through this lens, we identify critical limitations in existing evaluation metrics, which predominantly focus on pointwise faithfulness while neglecting contrastive reasoning and semantic granularity. To address these gaps, we propose four novel faithfulness metrics that systematically assess explanation quality across both RFxG dimensions. Our comprehensive evaluation framework spans ten state-of-the-art methods, 4 model architectures, and 3 datasets. By suggesting a shift from model-centric to user-intent-driven evaluation, our work provides both the conceptual foundation and practical tools necessary for developing explanations that are not only faithful to model behavior but also meaningfully aligned with human understanding.
Zitieren
APA
Elisha, Y., Cohen, S., Barkan, O., & Koenigstein, N. (2026). Rethinking Saliency Maps: A Cognitive Human Aligned Taxonomy and Evaluation Framework for Explanations. Proceedings of the AAAI Conference on Artificial Intelligence, 40(5), 3750-3758. https://doi.org/10.1609/aaai.v40i5.37375
Vancouver
Elisha Y, Cohen S, Barkan O, Koenigstein N. Rethinking Saliency Maps: A Cognitive Human Aligned Taxonomy and Evaluation Framework for Explanations. Proceedings of the AAAI Conference on Artificial Intelligence. 2026;40(5):3750-3758. doi:10.1609/aaai.v40i5.37375
BibTeX
@article{Elisha2026,
author = {Elisha, Yehonatan and Cohen, Seffi and Barkan, Oren and Koenigstein, Noam},
title = {Rethinking Saliency Maps: A Cognitive Human Aligned Taxonomy and Evaluation Framework for Explanations},
journal = {Proceedings of the AAAI Conference on Artificial Intelligence},
year = {2026},
volume = {40},
number = {5},
pages = {3750--3758},
doi = {10.1609/aaai.v40i5.37375}
}
EUVIMED