Título: Curvas Principais para a Seleção de Dados de Treinamento Neural com Grandes Volumes de Dados
Autores: Fernando Borges, José Seixas, Danton Ferreira
Resumo: In environments with big data problems, to make a smart data selection with the goal of training machine can be essential to reduce the computational demand of the application. This paper presents a method based on principal curves for data selection during the neural networks training in an experiment of particle collision with high events rate. The method used real data of collision and it accomplished 3 selection approaches through mapping of Euclidean distances in each event to the respective Principal Curve. Preliminary results in the classification of neural networks presented low differences using the selection method and considerable reduction in the training time.
Palavras-chave: Principal Curves; Data Selection; Big Data.
Código DOI: 10.21528/CBIC2019-21
Artigo em pdf: CBIC2019-21.pdf
Arquivo BibTeX: CBIC2019-21.bib