Saliency Maps Give a False Sense of Explanability to Image Classifiers: An Empirical Evaluation across Methods and Metrics

Hanwei Zhang, Felipe Torres Figueroa, Holger Hermanns
Proceedings of the 16th Asian Conference on Machine Learning, PMLR 260:479-494, 2025.

Abstract

The interpretability of deep neural networks (DNNs) has emerged as a crucial area of research, particularly in image classification tasks where decisions often lack transparency. Saliency maps have been widely used as a tool to decode the inner workings of these networks by highlighting regions of input images deemed most influential in the classification process. However, recent studies have revealed significant limitations and inconsistencies in the utility of saliency maps as explanations. This paper aims to systematically assess the shortcomings of saliency maps and explore alternative approaches to achieve more reliable and interpretable explanations for image classification models. We carry out a series of experiments to show that 1) the existing evaluation does not provide a fair nor meaningful comparison to the existing saliency maps; these evaluations have their implicit assumption and are not differentiable; 2) the saliency maps do not provide enough information on explaining the accuracy of network, the relationship between classes and the modification of the images.

Cite this Paper

BibTeX
@InProceedings{pmlr-v260-zhang25a, title = {{Saliency Maps Give a False Sense of Explanability to Image Classifiers}: {A}n Empirical Evaluation across Methods and Metrics}, author = {Zhang, Hanwei and Figueroa, Felipe Torres and Hermanns, Holger}, booktitle = {Proceedings of the 16th Asian Conference on Machine Learning}, pages = {479--494}, year = {2025}, editor = {Nguyen, Vu and Lin, Hsuan-Tien}, volume = {260}, series = {Proceedings of Machine Learning Research}, month = {05--08 Dec}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v260/main/assets/zhang25a/zhang25a.pdf}, url = {https://proceedings.mlr.press/v260/zhang25a.html}, abstract = {The interpretability of deep neural networks (DNNs) has emerged as a crucial area of research, particularly in image classification tasks where decisions often lack transparency. Saliency maps have been widely used as a tool to decode the inner workings of these networks by highlighting regions of input images deemed most influential in the classification process. However, recent studies have revealed significant limitations and inconsistencies in the utility of saliency maps as explanations. This paper aims to systematically assess the shortcomings of saliency maps and explore alternative approaches to achieve more reliable and interpretable explanations for image classification models. We carry out a series of experiments to show that 1) the existing evaluation does not provide a fair nor meaningful comparison to the existing saliency maps; these evaluations have their implicit assumption and are not differentiable; 2) the saliency maps do not provide enough information on explaining the accuracy of network, the relationship between classes and the modification of the images.} }
Endnote
%0 Conference Paper %T Saliency Maps Give a False Sense of Explanability to Image Classifiers: An Empirical Evaluation across Methods and Metrics %A Hanwei Zhang %A Felipe Torres Figueroa %A Holger Hermanns %B Proceedings of the 16th Asian Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2025 %E Vu Nguyen %E Hsuan-Tien Lin %F pmlr-v260-zhang25a %I PMLR %P 479--494 %U https://proceedings.mlr.press/v260/zhang25a.html %V 260 %X The interpretability of deep neural networks (DNNs) has emerged as a crucial area of research, particularly in image classification tasks where decisions often lack transparency. Saliency maps have been widely used as a tool to decode the inner workings of these networks by highlighting regions of input images deemed most influential in the classification process. However, recent studies have revealed significant limitations and inconsistencies in the utility of saliency maps as explanations. This paper aims to systematically assess the shortcomings of saliency maps and explore alternative approaches to achieve more reliable and interpretable explanations for image classification models. We carry out a series of experiments to show that 1) the existing evaluation does not provide a fair nor meaningful comparison to the existing saliency maps; these evaluations have their implicit assumption and are not differentiable; 2) the saliency maps do not provide enough information on explaining the accuracy of network, the relationship between classes and the modification of the images.
APA
Zhang, H., Figueroa, F.T. & Hermanns, H.. (2025). Saliency Maps Give a False Sense of Explanability to Image Classifiers: An Empirical Evaluation across Methods and Metrics. Proceedings of the 16th Asian Conference on Machine Learning, in Proceedings of Machine Learning Research 260:479-494 Available from https://proceedings.mlr.press/v260/zhang25a.html.

Related Material