DLR-Logo -> http://www.dlr.de
DLR Portal Home | Imprint | Privacy Policy | Contact | Deutsch
Fontsize: [-] Text [+]

Evaluation of the Explanatory Power Of Layer-wise Relevance Propagation using Adversarial Examples

Dieter, Tamara and Zisgen, Horst (2023) Evaluation of the Explanatory Power Of Layer-wise Relevance Propagation using Adversarial Examples. Neural Processing Letters. Springer Nature. doi: 10.1007/s11063-023-11166-8. ISSN 1370-4621.

[img] PDF - Published version

Official URL: https://link.springer.com/article/10.1007/s11063-023-11166-8


Approaches for visualizing and explaining the decision process of convolutional neural networks (CNNs) have recently received increasing attention. Particularly popular approaches are so-called saliency methods, which aim to assign a valence to each input pixel based on its importance and influence on the classification via saliency maps. In our paper, we contribute by a novel analyzing approach build on adversarial examples to investigate the explanatory power of saliency methods exemplified by layer-wise relevance propagation (LRP). Based on the hypothesis that distinct decisions, such as an image's classification and the classification of its corresponding adversarial examples, should yield to dissimilar saliency maps to provide transparent rationales, we break down relevance scores of images and corresponding adversarial examples and analyze them using a comprehensive statistical evaluation. It turns out that different relevance decomposition rules of LRP do not lead to clearly distinguishable saliency maps for images and corresponding adversarial examples, neither in terms of their contour lines, nor in terms of the statistical analysis.

Item URL in elib:https://elib.dlr.de/193232/
Document Type:Article
Title:Evaluation of the Explanatory Power Of Layer-wise Relevance Propagation using Adversarial Examples
AuthorsInstitution or Email of AuthorsAuthor's ORCID iDORCID Put Code
Dieter, TamaraUNSPECIFIEDhttps://orcid.org/0000-0001-9191-0170UNSPECIFIED
Zisgen, HorstUNSPECIFIEDhttps://orcid.org/0000-0003-4002-2012UNSPECIFIED
Date:10 March 2023
Journal or Publication Title:Neural Processing Letters
Refereed publication:Yes
Open Access:Yes
Gold Open Access:No
In ISI Web of Science:Yes
EditorsEmailEditor's ORCID iDORCID Put Code
Publisher:Springer Nature
Series Name:Neural Processing Letters
Keywords:Deep Learning, Layer-wise Relevance Propagation, Adversarial Examples, Explainable Artificial Intelligence, Saliency Maps
HGF - Research field:other
HGF - Program:other
HGF - Program Themes:other
DLR - Research area:no assignment
DLR - Program:no assignment
DLR - Research theme (Project):no assignment
Location: Rhein-Sieg-Kreis
Institutes and Institutions:Institute for the Protection of Terrestrial Infrastructures > Digital Twins of Infrastructures
Institute for the Protection of Terrestrial Infrastructures
Deposited By: Lenhard, Tamara
Deposited On:17 Mar 2023 09:25
Last Modified:24 Jul 2023 11:08

Repository Staff Only: item control page

Help & Contact
electronic library is running on EPrints 3.3.12
Website and database design: Copyright © German Aerospace Center (DLR). All rights reserved.