<div class="csl-bib-body">
<div class="csl-entry">Sietzen, S., Lechner, M., Borowski, J., Hasani, R., & Waldner, M. (2021). Interactive Analysis of CNN Robustness. <i>Computer Graphics Forum</i>, <i>40</i>(7), 253–264. https://doi.org/10.1111/cgf.14418</div>
</div>
-
dc.identifier.issn
0167-7055
-
dc.identifier.uri
http://hdl.handle.net/20.500.12708/55628
-
dc.description.abstract
While convolutional neural networks (CNNs) have found wide adoption as state-of-the-art models for image-related tasks, their predictions are often highly sensitive to small input perturbations, which the human vision is robust against. This paper presents Perturber, a web-based application that allows users to instantaneously explore how CNN activations and predictions evolve when a 3D input scene is interactively perturbed. Perturber offers a large variety of scene modifications, such as camera controls, lighting and shading effects, background modifications, object morphing, as well as adversarial attacks, to facilitate the discovery of potential vulnerabilities. Fine-tuned model versions can be directly compared for qualitative evaluation of their robustness. Case studies with machine learning experts have shown that Perturber helps users to quickly generate hypotheses about model vulnerabilities and to qualitatively compare model behavior. Using quantitative analyses, we could replicate users' insights with other CNN architectures and input images, yielding new insights about the vulnerability of adversarially trained models.
en
dc.language.iso
en
-
dc.publisher
WILEY
-
dc.relation.ispartof
Computer Graphics Forum
-
dc.subject
Computer Graphics and Computer-Aided Design
-
dc.title
Interactive Analysis of CNN Robustness
en
dc.type
Artikel
de
dc.type
Article
en
dc.description.startpage
253
-
dc.description.endpage
264
-
dc.type.category
Original Research Article
-
tuw.booktitle
Computer Graphics Forum
-
tuw.container.volume
40
-
tuw.container.issue
7
-
tuw.journal.peerreviewed
true
-
tuw.peerreviewed
true
-
tuw.book.ispartofseries
Computer Graphics Forum
-
tuw.researchTopic.id
I5
-
tuw.researchTopic.name
Visual Computing and Human-Centered Technology
-
tuw.researchTopic.value
100
-
dcterms.isPartOf.title
Computer Graphics Forum
-
tuw.publication.orgunit
E191-01 - Forschungsbereich Cyber-Physical Systems
-
tuw.publication.orgunit
E193-02 - Forschungsbereich Computer Graphics
-
tuw.publisher.doi
10.1111/cgf.14418
-
dc.identifier.eissn
1467-8659
-
dc.description.numberOfPages
12
-
tuw.author.orcid
0000-0002-6117-0076
-
tuw.author.orcid
0000-0002-3746-7152
-
tuw.author.orcid
0000-0002-9889-5222
-
tuw.author.orcid
0000-0003-1387-5132
-
wb.sci
true
-
wb.sciencebranch
Informatik
-
wb.sciencebranch.oefos
1020
-
wb.facultyfocus
Visual Computing and Human-Centered Technology (VC + HCT)
de
wb.facultyfocus
Visual Computing and Human-Centered Technology (VC + HCT)
en
wb.facultyfocus.faculty
E180
-
item.languageiso639-1
en
-
item.fulltext
no Fulltext
-
item.openairetype
research article
-
item.grantfulltext
restricted
-
item.openairecristype
http://purl.org/coar/resource_type/c_2df8fbb1
-
item.cerifentitytype
Publications
-
crisitem.author.dept
E191-01 - Forschungsbereich Cyber-Physical Systems
-
crisitem.author.dept
E191-01 - Forschungsbereich Cyber-Physical Systems
-
crisitem.author.dept
E193-02 - Forschungsbereich Computer Graphics
-
crisitem.author.orcid
0000-0002-9889-5222
-
crisitem.author.orcid
0000-0003-1387-5132
-
crisitem.author.parentorg
E191 - Institut für Computer Engineering
-
crisitem.author.parentorg
E191 - Institut für Computer Engineering
-
crisitem.author.parentorg
E193 - Institut für Visual Computing and Human-Centered Technology