Show simple item record

Theoretical evidence for adversarial robustness through randomization

dc.contributor.authorPinot, Rafaël
dc.contributor.authorMeunier, Laurent
dc.contributor.authorAraújo, Alexandre
dc.contributor.authorKashima, Hisashi
dc.contributor.authorYger, Florian
HAL ID: 17768
ORCID: 0000-0002-7182-8062
dc.contributor.authorGouy-Pailler, Cedric
HAL ID: 6827
ORCID: 0000-0003-1298-7845
dc.contributor.authorAtif, Jamal
HAL ID: 15689
dc.date.accessioned2020-10-23T12:12:47Z
dc.date.available2020-10-23T12:12:47Z
dc.date.issued2019
dc.identifier.urihttps://basepub.dauphine.fr/handle/123456789/21152
dc.language.isoenen
dc.subjectMachine Learning
dc.subject.ddc004en
dc.titleTheoretical evidence for adversarial robustness through randomization
dc.typeCommunication / Conférence
dc.description.abstractenThis paper investigates the theory of robustness against adversarial attacks. It focuses on the family of randomization techniques that consist in injecting noise in the network at inference time. These techniques have proven effective in many contexts, but lack theoretical arguments. We close this gap by presenting a theoretical analysis of these approaches, hence explaining why they perform well in practice. More precisely, we make two new contributions. The first one relates the randomization rate to robustness to adversarial attacks. This result applies for the general family of exponential distributions, and thus extends and unifies the previous approaches. The second contribution consists in devising a new upper bound on the adversarial generalization gap of randomized neural networks. We support our theoretical claims with a set of experiments.
dc.identifier.urlsitehttps://hal.archives-ouvertes.fr/hal-02892188
dc.subject.ddclabelInformatique généraleen
dc.relation.conftitle33rd Conference on Neural Information Processing Systems (NIPS 2019)
dc.relation.confdate2019-12
dc.relation.confcityVancouver
dc.relation.confcountryCANADA
dc.relation.forthcomingnonen
dc.description.ssrncandidatenon
dc.description.halcandidatenon
dc.description.readershiprecherche
dc.description.audienceInternational
dc.date.updated2021-01-12T15:15:08Z


Files in this item

FilesSizeFormatView

There are no files associated with this item.

This item appears in the following Collection(s)

Show simple item record