Multi-layer Perceptron and Pruning

Authors : Cyril VOYANT, Christophe PAOLİ, Marielaure NİVET, Gilles NOTTON, Alexis FOUİLLOY, Fabrice MOTTE

Pages : 1-6

View : 13 | Download : 10

Publication Date : 2017-08-22

Article Type : Research Paper

Abstract :A Multi-Layer Perceptron insert ignore into journalissuearticles values(MLP); defines a family of artificial neural networks often used in TS modeling and forecasting. Because of its “black box” aspect, many researchers refuse to use it. Moreover, the optimization insert ignore into journalissuearticles values(often based on the exhaustive approach where “all” configurations are tested); and learning phases of this artificial intelligence tool insert ignore into journalissuearticles values(often based on the Levenberg-Marquardt algorithm; LMA); are weaknesses of this approach insert ignore into journalissuearticles values(exhaustively and local minima);. These two tasks must be repeated depending on the knowledge of each new problem studied, making the process, long, laborious and not systematically robust. In this short communication, a pruning process is presented. This method allows, during the training phase, to carry out an inputs selecting method activating insert ignore into journalissuearticles values(or not); inter-nodes connections in order to verify if forecasting is improved. We propose to use iteratively the popular damped least-squares method to activate inputs and neurons. A first pass is applied to 10% of the learning sample to determine weights significantly different from 0 and delete other. Then a classical batch process based on LMA is used with the new MLP. The validation is done using 25 measured meteorological TS and cross-comparing the prediction results of the classical LMA and the 2-stage LMA.
Keywords : ANN, Forecasting, Time series, Meteorology