• Refine Query
  • Source
  • Publication year
  • to
  • Language
  • 1
  • Tagged with
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • About
  • The Global ETD Search service is a free service for researchers to find electronic theses and dissertations. This service is provided by the Networked Digital Library of Theses and Dissertations.
    Our metadata is collected from universities around the world. If you manage a university/consortium/country archive and want to be added, details can be found on the NDLTD website.
1

Classifica??o multirr?tulo com aprendizado semissupervisionado: uma an?lise multivis?o de dados

Assis, Mateus Silv?rio de 19 August 2016 (has links)
Submitted by Automa??o e Estat?stica (sst@bczm.ufrn.br) on 2017-02-20T21:14:32Z No. of bitstreams: 1 MateusSilverioDeAssis_DISSERT.pdf: 3929319 bytes, checksum: 7463541e5cc8c5aebedc5bd30d218bd4 (MD5) / Approved for entry into archive by Arlan Eloi Leite Silva (eloihistoriador@yahoo.com.br) on 2017-02-23T20:13:22Z (GMT) No. of bitstreams: 1 MateusSilverioDeAssis_DISSERT.pdf: 3929319 bytes, checksum: 7463541e5cc8c5aebedc5bd30d218bd4 (MD5) / Made available in DSpace on 2017-02-23T20:13:22Z (GMT). No. of bitstreams: 1 MateusSilverioDeAssis_DISSERT.pdf: 3929319 bytes, checksum: 7463541e5cc8c5aebedc5bd30d218bd4 (MD5) Previous issue date: 2016-08-19 / Ao longo dos ?ltimos anos, as t?cnicas computacionais aplicadas ao aprendizado de m?quina t?m sido divididas ou categorizadas quanto ao grau de supervis?o presente nos conjuntos de treinamentos e quanto ao n?mero de r?tulos presente no atributo classe. Dentro dessas divis?es, encontramos o aprendizado semissupervisionado, t?cnica que trabalha muito bem quando nem todos os r?tulos dos exemplos do conjunto de treinamento s?o conhecidos. Por outro lado, a classifica??o multirr?tulo, tamb?m est? presente nessas categorias e prop?e classificar exemplos quando estes est?o associados a um ou mais r?tulos. A combina??o dessas t?cnicas de aprendizado gera a classifica??o multirr?tulo semissupervisionado. Ainda nesse contexto, existem vertentes que trabalham com o aprendizado semissupervisionado para dados de vis?o ?nica e aprendizado semissupervisionado para dados de vis?o m?ltipla. Os algoritmos de aprendizado semissupervisionado para dados de vis?o m?ltipla tem como ideia b?sica a explora??o da discord?ncia entre as predi??es dos diferentes classificadores, sendo este um assunto pouco abordado em pesquisas. Nesse sentido, esse trabalho prop?e novos m?todos para classifica??o multirr?tulo semissupervisionado em uma abordagem para dados de vis?o m?ltipla, mostra os resultados de alguns experimentos realizados com esses novos m?todos e compara alguns desses resultados com resultados de experimentos utilizando m?todos j? existentes. / In the the last years, the computational techniques used for machine learning have been divided or categorized according to the degree of supervision that exists in these training?s set and according on the number of labels in this class attribute. Within these divisions, we find the semi-supervised learning, a technique that works well when nor all labels examples of the training set are known. In the other hand, the multi-label classification also is present in these categories and it proposes to classify examples when they are associated with one or more labels. The combination of these learning techniques generates the classification semi-supervised multi-label. Also in this context, there are sides that work with the semi-supervised learning for single vision and semisupervised learning data for multiple viewing data. The semi-supervised learning algorithms for multiple viewing data has the basic idea of the exploitation of disagreements between the predictions of different classifiers, which is a subject rarely addressed in research. Thus, this work proposes the use of semi-supervised learning for multi-label classification using an approach with multiple viewing data, showing the results of some experiments and comparing some results of experiments using the new methods with the results of experiments using existing methods.

Page generated in 0.089 seconds