Navegando por Palavras-chave "Transferência De Aprendizado"

Agora exibindo 1 - 1 de 1
  • Nenhuma Miniatura disponível
    Item
    Somente Metadadados
    Previsão De Preços De Commodities Agrícolas Utilizando Redes Neurais Artificiais Com Transferência De Aprendizado
    (Universidade Federal de São Paulo (UNIFESP), 2017-08-16) Goncalves, Cassio Doria [UNIFESP]; Melo, Vinicius Veloso De [UNIFESP]; Universidade Federal de São Paulo (UNIFESP)
    Transfer Learning Is A Strategy To Solve A Target Problem By Reusing The Learning Of A Source Problem That Has Already Been Solved. In This Work, We Propose To Use Transfer Learning On A Multilayer Perceptron Artificial Neural Network To Forecast Agricultural Commodities Prices. The Transfer Learning Approach Was Applied On Five Different Time Series. We Analyzed The Influence Of Similarity Between Source And Target Series In Order To Choose The Best Source. For Evaluating The Similarity, We Used The Dynamic Time Warping Distance Measure. The Transfer Learning Approach Provided A Significant Reduction Of The Computational Training Cost In Ninety-Five Percent Of A Total Of Five Hundred Test Cases. In Addition, The Transfer Learning Approach Improved The Forecast Quality In More Than Half Of The Cases. We Conclude, Therefore, That The Transfer Learning Approach Was Applied Successfully In Time Series Forecast.