• Login
    View Item 
    •   DSpace Home
    • BIBLIOTECA DIGITAL: Teses & Dissertações
    • 40001016034P5 Programa de Pós-Graduação em Informática
    • Dissertações
    • View Item
    •   DSpace Home
    • BIBLIOTECA DIGITAL: Teses & Dissertações
    • 40001016034P5 Programa de Pós-Graduação em Informática
    • Dissertações
    • View Item
    JavaScript is disabled for your browser. Some features of this site may not work without it.

    Transferência de conhecimento da transformada shapelet para classificação de séries temporais em redes neurais convolucionais

    Thumbnail
    View/Open
    R - D - ALEXANDRE FELIPE MULLER DE SOUZA.pdf (1.504Mb)
    Date
    2021
    Author
    Souza, Alexandre Felipe Müller de, 1985-
    Metadata
    Show full item record
    Subject
    Inteligência artificial
    Mineração de dados (Computação)
    Ciência da Computação
    xmlui.dri2xhtml.METS-1.0.item-type
    Dissertação Digital
    Abstract
    Resumo: O crescente acúmulo na aquisição de dados, por mais diversos dispositivos, torna necessário o uso de ferramentas automatizadas para a descoberta de padrões. Dentre estas ferramentas automatizadas, a classificação de séries temporais, com aplicações em inúmeras atividades, melhora continuamente devido às bases de dados de séries públicas que permitem um marco comparativo. Também entre as diversas formas de se classificar uma série temporal, duas são foco deste trabalho: a classificação com símbolos (em especial shapelets) e as redes neurais convolucionais. A primitiva shapelet é um descritor de subsequências representativas para uma classe. A extração prévia dessa representação pode melhorar a capacidade de classificação de uma rede neural convolucional através da transferência de conhecimento. Neste trabalho, mostra-se um experimento onde a extração destas representações, antes do início do aprendizado da rede neural, muda o comportamento destes classificadores para bases específicas e, consequentemente, representa um ganho de representação em diversas bases.
     
    Abstract: Nowadays the evolution of sensor data acquisition needs the use of automatic tools to discover patterns. One of those automatic tools is the time series classification, with widely possible applications, that improves continuously due to public datasets that becomes a benchmark in researches. Among many ways to classify time series two, in particular, are explored here: the symbolic (specially Shapelet primitive) and the neural network classification. The shapelet primitive is a descriptor of subsequences that represents a class. The previous extraction of this representation can improve the classification capacity in a convolutional neuronal network using transfer learning. Furthermore, is shown an experiment where symbolic extraction, before the network training, changes the result of a classifier for most bases and improves the classification accuracy.
     
    URI
    https://hdl.handle.net/1884/73589
    Collections
    • Dissertações [348]

    DSpace software copyright © 2002-2016  DuraSpace
    Contact Us | Send Feedback
    Theme by 
    Atmire NV
     

     

    Browse

    All of DSpaceCommunities & CollectionsBy Issue DateAuthorsTitlesSubjectsxmlui.ArtifactBrowser.Navigation.browse_typeThis CollectionBy Issue DateAuthorsTitlesSubjectsxmlui.ArtifactBrowser.Navigation.browse_type

    My Account

    LoginRegister

    Statistics

    View Usage Statistics

    DSpace software copyright © 2002-2016  DuraSpace
    Contact Us | Send Feedback
    Theme by 
    Atmire NV