Close
Metadata

1. Identity statement
Reference TypeConference Paper (Conference Proceedings)
Sitesibgrapi.sid.inpe.br
Holder Codeibi 8JMKD3MGPEW34M/46T9EHH
Identifier8JMKD3MGPEW34M/43992TE
Repositorysid.inpe.br/sibgrapi/2020/09.16.19.14
Last Update2020:09.16.19.14.22 (UTC) administrator
Metadata Repositorysid.inpe.br/sibgrapi/2020/09.16.19.14.22
Metadata Last Update2022:06.14.00.00.04 (UTC) administrator
DOI10.1109/SIBGRAPI51738.2020.00050
Citation KeyLayzaPedrTorr:2020:1tLaMa
Title1-to-N Large Margin Classifier
FormatOn-line
Year2020
Access Date2024, May 04
Number of Files1
Size449 KiB
2. Context
Author1 Layza, Jaime Rocca
2 Pedrini, Helio
3 Torres, Ricardo da Silva
Affiliation1 Institute of Computing, University of Campinas, Campinas, SP, Brazil, 13083-852
2 Institute of Computing, University of Campinas, Campinas, SP, Brazil, 13083-852
3 Department of ICT and Natural Sciences, Norwegian University of Science and Technology (NTNU)
EditorMusse, Soraia Raupp
Cesar Junior, Roberto Marcondes
Pelechano, Nuria
Wang, Zhangyang (Atlas)
e-Mail Addresshelio@ic.unicamp.br
Conference NameConference on Graphics, Patterns and Images, 33 (SIBGRAPI)
Conference LocationPorto de Galinhas (virtual)
Date7-10 Nov. 2020
PublisherIEEE Computer Society
Publisher CityLos Alamitos
Book TitleProceedings
Tertiary TypeFull Paper
History (UTC)2020-09-16 19:14:22 :: helio@ic.unicamp.br -> administrator ::
2022-06-14 00:00:04 :: administrator -> helio@ic.unicamp.br :: 2020
3. Content and structure
Is the master or a copy?is the master
Content Stagecompleted
Transferable1
Version Typefinaldraft
KeywordsLarge Margin Classifier
Noise Label Data
Adversarial Attacks
AbstractCross entropy with softmax is the standard loss function for classification in neural networks. However, this function can suffer from limitations on discriminative power, lack of generalization, and propensity to overfitting. In order to address these limitations, several approaches propose to enforce a margin on the top of the neural network specifically at the softmax function. In this work, we present a novel formulation that aims to produce generalization and noise label robustness not only by imposing a margin at the top of the neural network, but also by using the entire structure of the mini-batch data. Based on the distance used for SVM to obtain maximal margin, we propose a broader distance definition called 1-to-N distance and an approximated probability function as the basis for our proposed loss function. We perform empirical experimentation on MNIST, CIFAR-10, and ImageNet32 datasets to demonstrate that our loss function has better generalization and noise label robustness properties than the traditional cross entropy method, showing improvements in the following tasks: generalization robustness, robustness in noise label data, and robustness against adversarial examples attacks.
Arrangement 1urlib.net > SDLA > Fonds > SIBGRAPI 2020 > 1-to-N Large Margin...
Arrangement 2urlib.net > SDLA > Fonds > Full Index > 1-to-N Large Margin...
doc Directory Contentaccess
source Directory Contentthere are no files
agreement Directory Content
agreement.html 16/09/2020 16:14 1.2 KiB 
4. Conditions of access and use
data URLhttp://urlib.net/ibi/8JMKD3MGPEW34M/43992TE
zipped data URLhttp://urlib.net/zip/8JMKD3MGPEW34M/43992TE
Languageen
Target FilePID6615191.pdf
User Grouphelio@ic.unicamp.br
Visibilityshown
Update Permissionnot transferred
5. Allied materials
Mirror Repositorysid.inpe.br/banon/2001/03.30.15.38.24
Next Higher Units8JMKD3MGPEW34M/43G4L9S
8JMKD3MGPEW34M/4742MCS
Citing Item Listsid.inpe.br/sibgrapi/2020/10.28.20.46 3
Host Collectionsid.inpe.br/banon/2001/03.30.15.38
6. Notes
Empty Fieldsarchivingpolicy archivist area callnumber contenttype copyholder copyright creatorhistory descriptionlevel dissemination edition electronicmailaddress group isbn issn label lineage mark nextedition notes numberofvolumes orcid organization pages parameterlist parentrepositories previousedition previouslowerunit progress project readergroup readpermission resumeid rightsholder schedulinginformation secondarydate secondarykey secondarymark secondarytype serieseditor session shorttitle sponsor subject tertiarymark type url volume
7. Description control
e-Mail (login)helio@ic.unicamp.br
update 


Close