No Cover Image

Journal article 441 views 26 downloads

Jacobian norm with Selective Input Gradient Regularization for interpretable adversarial defense

Deyin Liu Orcid Logo, Lin Yuanbo Wu, Bo Li, Farid Boussaid, Mohammed Bennamoun, Xianghua Xie Orcid Logo, Chengwu Liang, Yuanbo Wu Orcid Logo

Pattern Recognition, Volume: 145, Start page: 109902

Swansea University Authors: Xianghua Xie Orcid Logo, Yuanbo Wu Orcid Logo

Check full text

DOI (Published version): 10.1016/j.patcog.2023.109902

Abstract

Deep neural networks (DNNs) can be easily deceived by imperceptible alterations known as adversarial examples. These examples can lead to misclassification, posing a significant threat to the reliability of deep learning systems in real-world applications. Adversarial training (AT) is a popular tech...

Full description

Published in: Pattern Recognition
ISSN: 0031-3203
Published: Elsevier BV 2024
Online Access: Check full text

URI: https://cronfa.swan.ac.uk/Record/cronfa64108
Tags: Add Tag
No Tags, Be the first to tag this record!
first_indexed 2023-08-23T08:43:05Z
last_indexed 2023-08-23T08:43:05Z
id cronfa64108
recordtype SURis
fullrecord <?xml version="1.0" encoding="utf-8"?><rfc1807 xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xmlns:xsd="http://www.w3.org/2001/XMLSchema"><bib-version>v2</bib-version><id>64108</id><entry>2023-08-23</entry><title>Jacobian norm with Selective Input Gradient Regularization for interpretable adversarial defense</title><swanseaauthors><author><sid>b334d40963c7a2f435f06d2c26c74e11</sid><ORCID>0000-0002-2701-8660</ORCID><firstname>Xianghua</firstname><surname>Xie</surname><name>Xianghua Xie</name><active>true</active><ethesisStudent>false</ethesisStudent></author><author><sid>205b1ac5a767e977bebb5d6afd770784</sid><ORCID>0000-0001-6119-058X</ORCID><firstname>Yuanbo</firstname><surname>Wu</surname><name>Yuanbo Wu</name><active>true</active><ethesisStudent>false</ethesisStudent></author></swanseaauthors><date>2023-08-23</date><deptcode>MACS</deptcode><abstract>Deep neural networks (DNNs) can be easily deceived by imperceptible alterations known as adversarial examples. These examples can lead to misclassification, posing a significant threat to the reliability of deep learning systems in real-world applications. Adversarial training (AT) is a popular technique used to enhance robustness by training models on a combination of corrupted and clean data. However, existing AT-based methods often struggle to handle transferred adversarial examples that can fool multiple defense models, thereby falling short of meeting the generalization requirements for real-world scenarios. Furthermore, AT typically fails to provide interpretable predictions, which are crucial for domain experts seeking to understand the behavior of DNNs. To overcome these challenges, we present a novel approach called Jacobian norm and Selective Input Gradient Regularization (J-SIGR). Our method leverages Jacobian normalization to improve robustness and introduces regularization of perturbation-based saliency maps, enabling interpretable predictions. By adopting J-SIGR, we achieve enhanced defense capabilities and promote high interpretability of DNNs. We evaluate the effectiveness of J-SIGR across various architectures by subjecting it to powerful adversarial attacks. Our experimental evaluations provide compelling evidence of the efficacy of J-SIGR against transferred adversarial attacks, while preserving interpretability. The project code can be found at https://github.com/Lywu-github/jJ-SIGR.git.</abstract><type>Journal Article</type><journal>Pattern Recognition</journal><volume>145</volume><journalNumber/><paginationStart>109902</paginationStart><paginationEnd/><publisher>Elsevier BV</publisher><placeOfPublication/><isbnPrint/><isbnElectronic/><issnPrint>0031-3203</issnPrint><issnElectronic/><keywords>Selective input gradient regularization, Jacobian normalization, Adversarial robustness</keywords><publishedDay>1</publishedDay><publishedMonth>1</publishedMonth><publishedYear>2024</publishedYear><publishedDate>2024-01-01</publishedDate><doi>10.1016/j.patcog.2023.109902</doi><url>http://dx.doi.org/10.1016/j.patcog.2023.109902</url><notes/><college>COLLEGE NANME</college><department>Mathematics and Computer Science School</department><CollegeCode>COLLEGE CODE</CollegeCode><DepartmentCode>MACS</DepartmentCode><institution>Swansea University</institution><apcterm/><funders>This work was partially supported by NSFC U19A2073 , 62002096, 62001394, 62372150, 62176086.</funders><projectreference/><lastEdited>2024-09-05T12:05:57.0584227</lastEdited><Created>2023-08-23T09:40:24.6634830</Created><path><level id="1">Faculty of Science and Engineering</level><level id="2">School of Mathematics and Computer Science - Computer Science</level></path><authors><author><firstname>Deyin</firstname><surname>Liu</surname><orcid>0000-0002-0371-9921</orcid><order>1</order></author><author><firstname>Lin Yuanbo</firstname><surname>Wu</surname><order>2</order></author><author><firstname>Bo</firstname><surname>Li</surname><order>3</order></author><author><firstname>Farid</firstname><surname>Boussaid</surname><order>4</order></author><author><firstname>Mohammed</firstname><surname>Bennamoun</surname><order>5</order></author><author><firstname>Xianghua</firstname><surname>Xie</surname><orcid>0000-0002-2701-8660</orcid><order>6</order></author><author><firstname>Chengwu</firstname><surname>Liang</surname><order>7</order></author><author><firstname>Yuanbo</firstname><surname>Wu</surname><orcid>0000-0001-6119-058X</orcid><order>8</order></author></authors><documents><document><filename>64108__28348__a9bdff416cd1485d9b740a80ec38d58a.pdf</filename><originalFilename>64108.pdf</originalFilename><uploaded>2023-08-23T09:42:26.4795898</uploaded><type>Output</type><contentLength>4714696</contentLength><contentType>application/pdf</contentType><version>Proof</version><cronfaStatus>true</cronfaStatus><embargoDate>2024-08-22T00:00:00.0000000</embargoDate><copyrightCorrect>false</copyrightCorrect><language>eng</language></document></documents><OutputDurs/></rfc1807>
spelling v2 64108 2023-08-23 Jacobian norm with Selective Input Gradient Regularization for interpretable adversarial defense b334d40963c7a2f435f06d2c26c74e11 0000-0002-2701-8660 Xianghua Xie Xianghua Xie true false 205b1ac5a767e977bebb5d6afd770784 0000-0001-6119-058X Yuanbo Wu Yuanbo Wu true false 2023-08-23 MACS Deep neural networks (DNNs) can be easily deceived by imperceptible alterations known as adversarial examples. These examples can lead to misclassification, posing a significant threat to the reliability of deep learning systems in real-world applications. Adversarial training (AT) is a popular technique used to enhance robustness by training models on a combination of corrupted and clean data. However, existing AT-based methods often struggle to handle transferred adversarial examples that can fool multiple defense models, thereby falling short of meeting the generalization requirements for real-world scenarios. Furthermore, AT typically fails to provide interpretable predictions, which are crucial for domain experts seeking to understand the behavior of DNNs. To overcome these challenges, we present a novel approach called Jacobian norm and Selective Input Gradient Regularization (J-SIGR). Our method leverages Jacobian normalization to improve robustness and introduces regularization of perturbation-based saliency maps, enabling interpretable predictions. By adopting J-SIGR, we achieve enhanced defense capabilities and promote high interpretability of DNNs. We evaluate the effectiveness of J-SIGR across various architectures by subjecting it to powerful adversarial attacks. Our experimental evaluations provide compelling evidence of the efficacy of J-SIGR against transferred adversarial attacks, while preserving interpretability. The project code can be found at https://github.com/Lywu-github/jJ-SIGR.git. Journal Article Pattern Recognition 145 109902 Elsevier BV 0031-3203 Selective input gradient regularization, Jacobian normalization, Adversarial robustness 1 1 2024 2024-01-01 10.1016/j.patcog.2023.109902 http://dx.doi.org/10.1016/j.patcog.2023.109902 COLLEGE NANME Mathematics and Computer Science School COLLEGE CODE MACS Swansea University This work was partially supported by NSFC U19A2073 , 62002096, 62001394, 62372150, 62176086. 2024-09-05T12:05:57.0584227 2023-08-23T09:40:24.6634830 Faculty of Science and Engineering School of Mathematics and Computer Science - Computer Science Deyin Liu 0000-0002-0371-9921 1 Lin Yuanbo Wu 2 Bo Li 3 Farid Boussaid 4 Mohammed Bennamoun 5 Xianghua Xie 0000-0002-2701-8660 6 Chengwu Liang 7 Yuanbo Wu 0000-0001-6119-058X 8 64108__28348__a9bdff416cd1485d9b740a80ec38d58a.pdf 64108.pdf 2023-08-23T09:42:26.4795898 Output 4714696 application/pdf Proof true 2024-08-22T00:00:00.0000000 false eng
title Jacobian norm with Selective Input Gradient Regularization for interpretable adversarial defense
spellingShingle Jacobian norm with Selective Input Gradient Regularization for interpretable adversarial defense
Xianghua Xie
Yuanbo Wu
title_short Jacobian norm with Selective Input Gradient Regularization for interpretable adversarial defense
title_full Jacobian norm with Selective Input Gradient Regularization for interpretable adversarial defense
title_fullStr Jacobian norm with Selective Input Gradient Regularization for interpretable adversarial defense
title_full_unstemmed Jacobian norm with Selective Input Gradient Regularization for interpretable adversarial defense
title_sort Jacobian norm with Selective Input Gradient Regularization for interpretable adversarial defense
author_id_str_mv b334d40963c7a2f435f06d2c26c74e11
205b1ac5a767e977bebb5d6afd770784
author_id_fullname_str_mv b334d40963c7a2f435f06d2c26c74e11_***_Xianghua Xie
205b1ac5a767e977bebb5d6afd770784_***_Yuanbo Wu
author Xianghua Xie
Yuanbo Wu
author2 Deyin Liu
Lin Yuanbo Wu
Bo Li
Farid Boussaid
Mohammed Bennamoun
Xianghua Xie
Chengwu Liang
Yuanbo Wu
format Journal article
container_title Pattern Recognition
container_volume 145
container_start_page 109902
publishDate 2024
institution Swansea University
issn 0031-3203
doi_str_mv 10.1016/j.patcog.2023.109902
publisher Elsevier BV
college_str Faculty of Science and Engineering
hierarchytype
hierarchy_top_id facultyofscienceandengineering
hierarchy_top_title Faculty of Science and Engineering
hierarchy_parent_id facultyofscienceandengineering
hierarchy_parent_title Faculty of Science and Engineering
department_str School of Mathematics and Computer Science - Computer Science{{{_:::_}}}Faculty of Science and Engineering{{{_:::_}}}School of Mathematics and Computer Science - Computer Science
url http://dx.doi.org/10.1016/j.patcog.2023.109902
document_store_str 1
active_str 0
description Deep neural networks (DNNs) can be easily deceived by imperceptible alterations known as adversarial examples. These examples can lead to misclassification, posing a significant threat to the reliability of deep learning systems in real-world applications. Adversarial training (AT) is a popular technique used to enhance robustness by training models on a combination of corrupted and clean data. However, existing AT-based methods often struggle to handle transferred adversarial examples that can fool multiple defense models, thereby falling short of meeting the generalization requirements for real-world scenarios. Furthermore, AT typically fails to provide interpretable predictions, which are crucial for domain experts seeking to understand the behavior of DNNs. To overcome these challenges, we present a novel approach called Jacobian norm and Selective Input Gradient Regularization (J-SIGR). Our method leverages Jacobian normalization to improve robustness and introduces regularization of perturbation-based saliency maps, enabling interpretable predictions. By adopting J-SIGR, we achieve enhanced defense capabilities and promote high interpretability of DNNs. We evaluate the effectiveness of J-SIGR across various architectures by subjecting it to powerful adversarial attacks. Our experimental evaluations provide compelling evidence of the efficacy of J-SIGR against transferred adversarial attacks, while preserving interpretability. The project code can be found at https://github.com/Lywu-github/jJ-SIGR.git.
published_date 2024-01-01T12:05:57Z
_version_ 1809353914411122688
score 11.035634

Similar Items