Comparação de desempenho entre os modelos neurais ágeis ELM e WiSARD

• Ano de defesa: 2017
• Autor(a) principal: Oliveira, Luiz Fernando dos Reis de
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal do Rio de Janeiro; Brasil; Instituto Alberto Luiz Coimbra de Pós-Graduação e Pesquisa de Engenharia; Programa de Pós-Graduação em Engenharia de Sistemas e Computação; UFRJ
• Palavras-chave em Português: Engenharia de Sistemas e Computação; Modelos neurais x estudo comparativo.; CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO
• Link de acesso: http://hdl.handle.net/11422/8707

Neural models are popular in machine learning. Agile neural models are a subset of this kind of models and are characterized by presenting a significantly faster training time, being applied mainly in online learning domains. Two examples of agile neural models are the Extreme Learning Machine (ELM), a single hidden layer feedforward neural network which synaptic weights do not need to be iteractively adjusted, and the Wilkes, Stonham and Aleksander Recognition Device (WiSARD), a weightless neural network model with multiple discriminators that use neurons based on RAM memory structures. In this work, a comparative study between ELM and WiSARD models is made, aiming to evaluate both models performance when applied to different datasets having different characteristics. The evaluation is made by comparing test accuracy, training and testing times metrics, as well as the amount of RAM memory consumed by the models.