Aprendizado de máquina baseado em tensores e suas aplicacções para floresta de caminhos ótimos

• Ano de defesa: 2015
• Autor(a) principal: Lopes, Ricardo Ricci [UNESP]
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Estadual Paulista (Unesp)
• Palavras-chave em Português: Computação - Matematica; Processamento de imagens - Tecnicas digitais; Reconhecimento de padrões; Álgebra tensorial; Aprendizado do computador; Floresta de caminhos ótimos; Computer science Mathematics
• Link de acesso: http://hdl.handle.net/11449/136657; http://www.athena.biblioteca.unesp.br/exlibris/bd/cathedra/24-03-2016/000859943.pdf

Machine learning techniques usually learn some decision surface that separates samples from di erent classes by means of their vectorial representation. However, there exist many applications that might lose important information that are strongly related to the data itself. Additionally, such information has gained importance with the popularity of high-dimensional datasets. As such, works based on Mathematics and Physics, where curvature-based space representations have been used in several application, have gained attention by the machine learning community. Such representations are based on tensors, which keep the original structure of the data, as well as they allow us to use manifolds in curvature-based spaces. This master's dissertation presents a review of the literature with respect to tensor-based machine learning techniques, as well as a brief review about multilinear algebra. We also evaluate the performance of the Optimum-Path Forest classi er (OPF) in tensor-oriented spaces by means of the Multilinear Principal Component Analysis, as well as its comparison against with other related techniques is also performed. It is shown OPF can bene t from such feature space representation in some situations