Please use this identifier to cite or link to this item:
Type: Tese de Doutorado
Title: Distance-based clustering methods for large datasets
Authors: Gustavo Rodrigues Lacerda Silva
First Advisor: Antonio de Padua Braga
First Referee: Rodney Rezende Saldanha
Second Referee: Eduardo Mazoni Andrade Marcal Mendes
Third Referee: Luis Enrique Zarate Galvez
metadata.dc.contributor.referee4: Douglas Alexandre Gomes Vieira
Abstract: Este trabalho apresenta uma metodologia direcionada a problemas de agrupamentos com grandes volumes de dados. O objetivo ´e projetar algoritmos que tenham a capacidade de processar grandes volumes de dados sem a perda de qualidade do agrupamento. Dois novos m´etodos de agrupamento sao propostos. O primeiro ´e o m´etodo de agrupamento GPIC, que realiza tanto o c´alculo da matriz de afinidades quanto dos autovetores com o aux´lio de Unidades de Processamento Gr´afico GPUs, do ingles Graphics Processing Unit. O segundo m´etodo, denominado bdrFCM, reduz o volume de dados utilizando como princ´pio b´asico a borda dos agrupamentos resultantes. Resultados encontrados com bases de dados sint´eticas e reais demonstram que as abordagens propostas por este trabalho conseguem processar grande quantidade de dados em tempo menor e reduzir o volume de dados, mantendo a qualidade do agrupamento
Abstract: This PhD dissertation presents a methodology focused on clustering problems with large data volumes. The goal is to design algorithms that can process large volumes of data without loss of clustering quality. Specifically, this Doctoral dissertation presents two novel, fast and scalable distance-based clustering algorithms well suited to analyse large datasets. The first one is the GPIC clustering method, which performs the calculation of the anity matrix and the eigenvectors with the support of the Graphics Processing Unit - GPU. The second method, called bdrFCM, reduces the volume of data using the border of the Fuzzy c-means cluster results as a fundamental principle. Results found with synthetic and real datasets demonstrate that the approaches proposed by this work can process a significant amount of data in less time and reduce the volume of data, whilst maintaining the quality of the clustering result
Subject: Algoritmos
Engenharia elétrica
language: Inglês
Publisher: Universidade Federal de Minas Gerais
Publisher Initials: UFMG
Rights: Acesso Aberto
Issue Date: 30-Jul-2018
Appears in Collections:Teses de Doutorado

Files in This Item:
File Description SizeFormat 
85d.pdf8.93 MBAdobe PDFView/Open

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.