Mixed Radial Basis Function Neural Network Training Using Genetic Algorithm

• Published in: Neural processing letters Vol. 55; no. 8; pp. 10569 - 10587
• Main Authors: Elansari, Taoufyq, Ouanan, Mohammed, Bourray, Hamid
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.12.2023; Springer Nature B.V
• Subjects: Approximation; Artificial Intelligence; Artificial neural networks; Complex Systems; Computational Intelligence; Computer Science; Constraint modelling; Genetic algorithms; Interpolation; Mathematical analysis; Methods; Neural networks; Neurons; Optimization; Radial basis function; Saturn; Training; Variables; Mixed-variable optimization; Radial basis function neural networks; Genetic algorithm; Universal approximation
• ISSN: 1370-4621; 1573-773X
• DOI: 10.1007/s11063-023-11339-5

The Radial Basis Function Neural Network (RBFNN) is a feedforward artificial neural network employing radial basis functions as activation functions in the hidden layer. The output of the RBFNN is a linear combination of the outputs from the hidden layer. This paper present a Mixed Radial Basis Function Neural Network (MRBFNN) training using Genetic Algorithm (GA). The choice of the type of Radial Basis Functions (RBFs) utilized in each hidden layer neuron has a significant impact on convergence, interpolation and performance. In this work, the authors are introducing a new approach to optimizing the choice of radial basis functions, centers, radius and weights of the output layer. We model in terms of mixed-variable optimization problems with linear constraints. To solve this model we will use an approach based on the genetic algorithm, allows us to determine the types of RBF to use in the hidden layer and the optimal weight of the output layer which gives us a good generalization. The results numerically demonstrate the performance of the theoretic results presented in this paper, as well as the benefits of the new modeling.