To tune or not to tune: Recommending when to adjust SVM hyper-parameters via meta-learning

Página do item simplificado
dc.contributor.authorMantovani, Rafael G.
dc.contributor.authorRossi, André L. D. [UNESP]
dc.contributor.authorVanschoren, Joaquin
dc.contributor.authorBischl, Bernd
dc.contributor.authorCarvalho, André C.P.L.F.
dc.contributor.institutionUniversidade de São Paulo (USP)
dc.contributor.institutionUniversidade Estadual Paulista (Unesp)
dc.contributor.institutionEindhoven University of Technology (TU/e)
dc.contributor.institutionLudwig-Maximilians-University Munich
dc.date.accessioned2018-12-11T16:39:39Z
dc.date.available2018-12-11T16:39:39Z
dc.date.issued2015-09-28
dc.description.abstractMany classification algorithms, such as Neural Networks and Support Vector Machines, have a range of hyper-parameters that may strongly affect the predictive performance of the models induced by them. Hence, it is recommended to define the values of these hyper-parameters using optimization techniques. While these techniques usually converge to a good set of values, they typically have a high computational cost, because many candidate sets of values are evaluated during the optimization process. It is often not clear whether this will result in parameter settings that are significantly better than the default settings. When training time is limited, it may help to know when these parameters should definitely be tuned. In this study, we use meta-learning to predict when optimization techniques are expected to lead to models whose predictive performance is better than those obtained by using default parameter settings. Hence, we can choose to employ optimization techniques only when they are expected to improve performance, thus reducing the overall computational cost. We evaluate these meta-learning techniques on more than one hundred data sets. The experimental results show that it is possible to accurately predict when optimization techniques should be used instead of default values suggested by some machine learning libraries.en
dc.description.affiliationUniversidade de São Paulo (USP)
dc.description.affiliationUniversidade Estadual Paulista (UNESP)
dc.description.affiliationEindhoven University of Technology (TU/e)
dc.description.affiliationLudwig-Maximilians-University Munich
dc.description.affiliationUnespUniversidade Estadual Paulista (UNESP)
dc.identifierhttp://dx.doi.org/10.1109/IJCNN.2015.7280644
dc.identifier.citationProceedings of the International Joint Conference on Neural Networks, v. 2015-September.
dc.identifier.doi10.1109/IJCNN.2015.7280644
dc.identifier.scopus2-s2.0-84944265251
dc.identifier.urihttp://hdl.handle.net/11449/168078
dc.language.isoeng
dc.relation.ispartofProceedings of the International Joint Conference on Neural Networks
dc.rights.accessRightsAcesso aberto
dc.sourceScopus
dc.subjectComputational modeling
dc.subjectNickel
dc.subjectNiobium
dc.subjectOptimization
dc.subjectRadio frequency
dc.subjectSupport vector machines
dc.subjectTraining
dc.titleTo tune or not to tune: Recommending when to adjust SVM hyper-parameters via meta-learningen
dc.typeTrabalho apresentado em evento
unesp.author.lattes5604829226181486[2]
unesp.author.orcid0000-0001-6388-7479[2]

Arquivos

Coleções

Artigos