Show simple item record

dc.contributor.authorTsipras, Dimitris
dc.contributor.authorSanturkar, Shibani (Shibani Vinay)
dc.contributor.authorEngstrom, Logan G.
dc.contributor.authorTurner, Alexander M.
dc.contributor.authorMadry, Aleksander
dc.date.accessioned2021-03-05T14:59:33Z
dc.date.available2021-03-05T14:59:33Z
dc.date.issued2019-04
dc.identifier.urihttps://hdl.handle.net/1721.1/130090
dc.description.abstractWe show that there exists an inherent tension between the goal of adversarial robustness and that of standard generalization. Specifically, training robust models may not only be more resource-consuming, but also lead to a reduction of standard accuracy. We demonstrate that this trade-off between the standard accuracy of a model and its robustness to adversarial perturbations provably exists even in a fairly simple and natural setting. These findings also corroborate a similar phenomenon observed in practice. Further, we argue that this phenomenon is a consequence of robust classifiers learning fundamentally different feature representations than standard classifiers. These differences, in particular, seem to result in unexpected benefits: the features learned by robust models tend to align better with salient data characteristics and human perception.en_US
dc.description.sponsorshipNational Science Foundation (U.S.) (Grants S-1447786, IIS-1607189, CCF-1563880,CCF-1553428)en_US
dc.language.isoen
dc.publisherICLRen_US
dc.relation.isversionofhttps://openreview.net/forum?id=SyxAb30cY7en_US
dc.rightsCreative Commons Attribution-Noncommercial-Share Alikeen_US
dc.rights.urihttp://creativecommons.org/licenses/by-nc-sa/4.0/en_US
dc.sourcearXiven_US
dc.titleRobustness may be at odds with accuracyen_US
dc.typeArticleen_US
dc.identifier.citationTsipras, Dimitris et al. “Robustness may be at odds with accuracy.” Paper presented at 7th International Conference on Learning Representations, ICLR 2019, New Orleans, Louisiana, May 6 - 9, 2019, ICLR © 2019 The Author(s)en_US
dc.contributor.departmentMassachusetts Institute of Technology. Department of Electrical Engineering and Computer Scienceen_US
dc.relation.journal7th International Conference on Learning Representations, ICLR 2019en_US
dc.eprint.versionAuthor's final manuscripten_US
dc.type.urihttp://purl.org/eprint/type/ConferencePaperen_US
eprint.statushttp://purl.org/eprint/status/NonPeerRevieweden_US
dc.date.updated2021-02-05T15:05:55Z
dspace.orderedauthorsTsipras, D; Santurkar, S; Engstrom, L; Turner, A; Madry, Aen_US
dspace.date.submission2021-02-05T15:06:42Z
mit.journal.volume2019en_US
mit.licenseOPEN_ACCESS_POLICY


Files in this item

Thumbnail

This item appears in the following Collection(s)

Show simple item record