Cardiff University | Prifysgol Caerdydd ORCA
Online Research @ Cardiff 
WelshClear Cookie - decide language by browser settings

Sanity checks for saliency metrics

Tomsett, Richard, Harborne, Daniel, Chakraborty, Supriyo, Gurram, Prudhvi and Preece, Alun D. ORCID: https://orcid.org/0000-0003-0349-9057 2020. Sanity checks for saliency metrics. Presented at: AAAI-20: 34th AAAI Conference on Artificial Intelligence, New York City, NY, USA, 7-12 February 2020. Proceedings of the 34th AAAI Conference on Artificial Intelligence. , vol.34(04) AAAI Press, pp. 6021-6029. 10.1609/aaai.v34i04.6064

[thumbnail of AAAI-TomsettR.8794.pdf]
Preview
PDF - Accepted Post-Print Version
Download (1MB) | Preview

Abstract

Saliency maps are a popular approach to creating post-hoc explanations of image classifier outputs. These methods produce estimates of the relevance of each pixel to the classification output score, which can be displayed as a saliency map that highlights important pixels. Despite a proliferation of such methods, little effort has been made to quantify how good these saliency maps are at capturing the true relevance of the pixels to the classifier output (i.e. their “fidelity”). We therefore investigate existing metrics for evaluating the fidelity of saliency methods (i.e. saliency metrics). We find that there is little consistency in the literature in how such metrics are calculated, and show that such inconsistencies can have a significant effect on the measured fidelity. Further, we apply measures of reliability developed in the psychometric testing literature to assess the consistency of saliency metrics when applied to individual saliency maps. Our results show that saliency metrics can be statistically unreliable and inconsistent, indicating that comparative rankings between saliency methods generated using such metrics can be untrustworthy.

Item Type: Conference or Workshop Item (Paper)
Date Type: Publication
Status: Published
Schools: Schools > Computer Science & Informatics
Research Institutes & Centres > Crime and Security Research Institute (CSURI)
Publisher: AAAI Press
ISBN: 9781577358350
Related URLs:
Date of First Compliant Deposit: 20 December 2019
Date of Acceptance: 11 November 2019
Last Modified: 24 Sep 2025 10:00
URI: https://orca.cardiff.ac.uk/id/eprint/127677

Citation Data

Cited 22 times in Scopus. View in Scopus. Powered By Scopus® Data

Actions (repository staff only)

Edit Item Edit Item

Downloads

Downloads per month over past year

View more statistics