Engineering Applications of Artificial Intelligence
| Hi-index | 0.00 |
In the present work, a constructive learning algorithmis employed to design an optimal one-hidden neuralnetwork structure that best approximates a givenmapping. The method determines not only the optimalnumber of hidden neurons but also the best activationfunction for each node. Here, the projection pursuittechnique is applied in association with the optimizationof the solvability condition, giving rise to a moreefficient and accurate computational learning algorithm.As each activation function of a hidden neuron isoptimally defined for every approximation problem,better rates of convergence are achieved. Since thetraining process operates the hidden neuronsindividually, a pertinent activation function employingHermite polynomials can be iteratively developed foreach neuron as a function of the learning set. Theproposed constructive learning algorithm wassuccessfully applied to identify a large-scale multivariateprocess, providing a multivariable model that was ableto describe the complex process dynamics, even in long-rangehorizon predictions.