Show simple item record

dc.contributor.authorAlfarra, Motasem
dc.contributor.authorPérez, Juan C.
dc.contributor.authorBibi, Adel
dc.contributor.authorThabet, Ali Kassem
dc.contributor.authorArbeláez, Pablo
dc.contributor.authorGhanem, Bernard
dc.date.accessioned2020-12-01T11:33:05Z
dc.date.available2020-12-01T11:33:05Z
dc.date.issued2020-06-13
dc.identifier.urihttp://hdl.handle.net/10754/666190
dc.description.abstractThis paper studies how encouraging semantically-aligned features during deep neural network training can increase network robustness. Recent works observed that Adversarial Training leads to robust models, whose learnt features appear to correlate with human perception. Inspired by this connection from robustness to semantics, we study the complementary connection: from semantics to robustness. To do so, we provide a tight robustness certificate for distance-based classification models (clustering-based classifiers), which we leverage to propose ClusTR (Clustering Training for Robustness), a clustering-based and adversary-free training framework to learn robust models. Interestingly, ClusTR outperforms adversarially-trained networks by up to 4\% under strong PGD attacks. Moreover, it can be equipped with simple and fast adversarial training to improve the current state-of-the-art in robustness by 16\%-29\% on CIFAR10, SVHN, and CIFAR100.
dc.publisherarXiv
dc.relation.urlhttps://arxiv.org/pdf/2006.07682
dc.rightsArchived with thanks to arXiv
dc.titleClusTR: Clustering Training for Robustness
dc.typePreprint
dc.contributor.departmentElectrical Engineering
dc.contributor.departmentComputer, Electrical and Mathematical Sciences and Engineering (CEMSE) Division
dc.contributor.departmentElectrical Engineering Program
dc.contributor.departmentVisual Computing Center (VCC)
dc.contributor.departmentKAUST.
dc.eprint.versionPre-print
dc.contributor.institutionUniversidad de los Andes.
dc.identifier.arxivid2006.07682
kaust.personAlfarra, Motasem
kaust.personBibi, Adel
kaust.personThabet, Ali Kassem
kaust.personArbeláez, Pablo
kaust.personGhanem, Bernard
refterms.dateFOA2020-12-01T11:33:31Z


Files in this item

Thumbnail
Name:
Preprintfile1.pdf
Size:
1.229Mb
Format:
PDF
Description:
Pre-print

This item appears in the following Collection(s)

Show simple item record