Improving the Multilayer Perceptron Learning by Using a Method to Calculate the Initial Weights with the Similarity Quality Measure Based on Fuzzy Sets and Particle Swarms

• Lenniet Coello ^[1] ; Yumilka Fernandez ^[1] ; Yaima Filiberto ^[1] ; Rafael Bello ^[2]
  1. [1] Universidad de Camagüey

     Universidad de Camagüey

     Cuba

     
  2. [2] Universidad Central de Las Villas

     Universidad Central de Las Villas

     Cuba

     
• Localización: Computación y Sistemas (CyS), ISSN 1405-5546, ISSN-e 2007-9737, Vol. 19, Nº. 2, 2015, págs. 309-320
• Idioma: inglés
• Enlaces
  • Texto completo
• Resumen

  • The most widely used neural network model is Multilayer Perceptron (MLP), in which training of the connection weights is normally completed by a Back Propagation learning algorithm. Good initial values of weights bear a fast convergence and a better generalization capability even with simple gradient-based error minimization techniques. This work presents a method to calculate the initial weights in order to train the Multilayer Perceptron Model. The method named PSO+RST+FUZZY is based on the similarity quality measure proposed within the framework of the extended Rough Set Theory that employs fuzzy sets to characterize the domain of similarity thresholds. Sensitivity of BP to initial weights with PSO+RST+FUZZY was studied experimentally, showing better performance than other methods used to calculate feature weights.