Busca avançada
Ano de início
Entree
(Referência obtida automaticamente do Web of Science, por meio da informação sobre o financiamento pela FAPESP e o número do processo correspondente, incluída na publicação pelos autores.)

LoCH: A neighborhood-based multidimensional projection technique for high-dimensional sparse spaces

Texto completo
Autor(es):
Fadel, Samuel G. [1] ; Fatore, Francisco M. [1] ; Duarte, Felipe S. L. G. [1] ; Paulovich, Fernando V. [1]
Número total de Autores: 4
Afiliação do(s) autor(es):
[1] Univ Sao Paulo, Inst Ciencias Matemat & Comp, Sao Carlos, SP - Brazil
Número total de Afiliações: 1
Tipo de documento: Artigo Científico
Fonte: Neurocomputing; v. 150, n. B, SI, p. 546-556, FEB 20 2015.
Citações Web of Science: 9
Resumo

On the last few years multidimensional projection techniques have advanced towards defining faster and user-centered approaches. However, most of existing methods are designed as generic tools without considering particular features of the data under processing, such as the distance distribution when the data is embedded into a certain metric space. In this paper we split the projection techniques into two groups, global and local techniques, conduct an analysis of them, and present a novel local technique specially designed for projecting heavy tail distance distributions, such as the one produced by high-dimensional sparse spaces. This novel approach, called Local Convex Hull (LoCH), relies on an iterative process that seeks to place each point close to the convex hull of its nearest neighbors. The accuracy, in terms of neighborhood preservation, is confirmed by a set of comparisons and tests, showing that LoCH is capable of successfully segregating groups of similar instances embedded in high-dimensional sparse spaces and of defining the borders between them, significantly better than most projection techniques. (C) 2014 Elsevier B.V. All rights reserved. (AU)

Processo FAPESP: 11/22749-8 - Desafios em visualização exploratória de dados multidimensionais: novos paradigmas, escalabilidade e aplicações
Beneficiário:Luis Gustavo Nonato
Linha de fomento: Auxílio à Pesquisa - Temático
Processo FAPESP: 13/08665-1 - Desenvolvendo técnicas de projeção multi-dimensional para dispositivos móveis
Beneficiário:Samuel Gomes Fadel
Linha de fomento: Bolsas no Brasil - Iniciação Científica