• Entrar
    Ver item 
    •   Página inicial
    • BIBLIOTECA DIGITAL: Teses & Dissertações
    • 40001016041P1 Programa de Pós-Graduação em Matemática
    • Dissertações
    • Ver item
    •   Página inicial
    • BIBLIOTECA DIGITAL: Teses & Dissertações
    • 40001016041P1 Programa de Pós-Graduação em Matemática
    • Dissertações
    • Ver item
    JavaScript is disabled for your browser. Some features of this site may not work without it.

    Análise teórica de máquinas de vetores suporte e aplicação a classificação de caracteres

    Thumbnail
    Visualizar/Abrir
    R - D - EVELIN HERINGER MANOEL KRULIKOVSKI.pdf (2.677Mb)
    Data
    2017
    Autor
    Krulikovski, Evelin Heringer Manoel
    Metadata
    Mostrar registro completo
    Resumo
    Resumo: O objetivo geral deste trabalho foi realizar um estudo te.rico e uma pequena aplica..o sobre SVM, que inclui relatar justificativas para o uso de tal técnica e exibir sua interpretação geométrica e perspectiva analítica. Para aplicar a técnica em problemas de classificação, buscamos fundamentar matematicamente sua utilização, visto que envolve um problema de programação quadrática, convexa e com restrições. Para a análise da técnica, utilizamos a teoria de dualidade Lagrangiana, que notamos facilitar os cálculos e a análise das soluções. Alem disso, reescrevemos resultados que usam ponto de sela, sem precisar deste conceito. Estabelecemos algumas implicações e exibimos alguns contraexemplos, para mostrar que certos resultados decorrentes da técnica SVM encontrados na literatura não são precisos. Foram feitas algumas comparações, para analisar os diferentes parâmetros da função Kernel Gaussiana, usada para resolver o problema quando não for possível encontrar uma função de decisão no espaço de entrada. Verificamos que a eficiência da técnica depende da escolha do parâmetro de regulariza.ao, da função Kernel e seus respectivos parâmetros. Tal técnica foi testada sobre um banco de dados criado artificialmente e composto de imagens de caracteres. Para a implementação computacional, usamos a interface do programa Algencan e algumas funções próprias do Matlab.
     
    Abstract: The general objective of this work was to perform a theoretical study and a small application about SVM, which includes reporting justifications for the use of such technique and showing its geometric interpretation and analytical perspective. In order to apply the technique to classification problems, we seek to base its use mathematically, since it involves a quadratic, convex and constrained programming problem. For the analysis of the technique, we use the theory of Lagrangian duality, which we noticed to facilitate the calculations and the analysis of the solutions. In addition, we rewrite results using a saddle point, without needing this concept. We have established some implications and have shown some counterexamples to show that certain results from the SVM technique found in the literature are not accurate. Some comparisons have been made to analyze the different parameters of the Gaussian kernel function used to solve the problem when it is not possible to find a decision function in the input space. We have verified that the efficiency of the technique depends on the choice of the regularization parameter, the kernel function and its parameters. This technique was tested on an artificially created database composed of character images. For the computational implementation, we used the Algencan program interface and some of Matlab's own functions.
     
    URI
    https://hdl.handle.net/1884/52189
    Collections
    • Dissertações [60]

    DSpace software copyright © 2002-2022  LYRASIS
    Entre em contato | Deixe sua opinião
    Theme by 
    Atmire NV
     

     

    Navegar

    Todo o repositórioComunidades e ColeçõesPor data do documentoAutoresTítulosAssuntosTipoEsta coleçãoPor data do documentoAutoresTítulosAssuntosTipo

    Minha conta

    EntrarCadastro

    Estatística

    Ver as estatísticas de uso

    DSpace software copyright © 2002-2022  LYRASIS
    Entre em contato | Deixe sua opinião
    Theme by 
    Atmire NV