A self-organizing radial basis function network combined with ART II

To obtain good network performance RBF network needs more careful design consideration. The selection of several parameter values such as the number and centers of the radial basis functions must be considered carefully, since they critically affect its performance. We propose a new RBF network architecture, which can recruit neurons automatically to effectively find out appropriate centers best reflecting the characteristics of the input pattern. A self-organizing network, adaptive resonance theory (ART) II is combined with conventional RBF to achieve this. To demonstrate the performance of the proposed network and previously stated effect of network parameters, general problem of function estimation is treated. The representation problem of continuous functions defined over 2D input space is solved. The results obtained from the simulations show that the proposed RBF network yields satisfactory performance in terms of convergence and accuracy compared with those obtained by conventional multilayer perceptron network