Logotipo do repositório
 

Publicação:
Fine-Tuning Infinity Restricted Boltzmann Machines

dc.contributor.authorPassos, L. A.
dc.contributor.authorPapa, J. P. [UNESP]
dc.contributor.authorIEEE
dc.contributor.institutionUniversidade Federal de São Carlos (UFSCar)
dc.contributor.institutionUniversidade Estadual Paulista (Unesp)
dc.date.accessioned2018-11-26T17:48:13Z
dc.date.available2018-11-26T17:48:13Z
dc.date.issued2017-01-01
dc.description.abstractRestricted Boltzmann Machines (RBMs) have received special attention in the last decade due to their outstanding results in number of applications, such as face and human motion recognition, and collaborative filtering, among others. However, one of the main concerns about RBMs is related to the number of hidden units, which is application-dependent. Infinite RBM (iRBM) was proposed as an alternative to the regular RBM, where the number of units in the hidden layer grows as long as it is necessary, dropping out the need for selecting a proper number of hidden units. However, a less sensitive regularization parameter is introduced as well. This paper proposes to fine-tune iRBM hyper-parameters by means of meta-heuristic techniques such as Particle Swarm Optimization, Bat Algorithm, Cuckoo Search, and the Firefly Algorithm. The proposed approach is validated in the context of binary image reconstruction over two well-known datasets. Furthermore, the experimental results compare the robustness of the iRBM against the RBM and Ordered RBM (oRBM) using two different learning algorithms, showing the suitability in using meta-heuristics for hyper-parameter fine-tuning in RBM-based models.en
dc.description.affiliationUniv Fed Sao Carlos, Dept Comp, Sao Carlos, SP, Brazil
dc.description.affiliationSao Paulo State Univ, Dept Comp, Bauru, Brazil
dc.description.affiliationUnespSao Paulo State Univ, Dept Comp, Bauru, Brazil
dc.description.sponsorshipFundação de Amparo à Pesquisa do Estado de São Paulo (FAPESP)
dc.description.sponsorshipCoordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES)
dc.description.sponsorshipConselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq)
dc.description.sponsorshipIdFAPESP: 2014/12236-1
dc.description.sponsorshipIdFAPESP: 2016/19403-6
dc.description.sponsorshipIdCNPq: 306166/2014-3
dc.format.extent63-70
dc.identifierhttp://dx.doi.org/10.1109/SIBGRAPI.2017.15
dc.identifier.citation2017 30th Sibgrapi Conference On Graphics, Patterns And Images (sibgrapi). New York: Ieee, p. 63-70, 2017.
dc.identifier.doi10.1109/SIBGRAPI.2017.15
dc.identifier.issn1530-1834
dc.identifier.urihttp://hdl.handle.net/11449/163864
dc.identifier.wosWOS:000425243500009
dc.language.isoeng
dc.publisherIeee
dc.relation.ispartof2017 30th Sibgrapi Conference On Graphics, Patterns And Images (sibgrapi)
dc.rights.accessRightsAcesso aberto
dc.sourceWeb of Science
dc.subjectDeep Learning
dc.subjectInfinity Restricted Boltzmann Machines
dc.subjectMeta-heuristics
dc.titleFine-Tuning Infinity Restricted Boltzmann Machinesen
dc.typeTrabalho apresentado em evento
dcterms.licensehttp://www.ieee.org/publications_standards/publications/rights/rights_policies.html
dcterms.rightsHolderIeee
dspace.entity.typePublication
unesp.campusUniversidade Estadual Paulista (UNESP), Faculdade de Ciências, Baurupt
unesp.departmentComputação - FCpt

Arquivos