Determining the centers of radial basis probabilistic neural networks by recursive orthogonal least square algorithms

In this paper, we adopt a recursive orthogonal least squares algorithm (ROLSA) to train radial basis probabilistic neural networks (RBPNN) and select the corresponding hidden centers from the training samples. The ROLSA is first used to recursively find the weights between the second hidden layer and the output layer of the RBPNN. Then, the basic principle to select the hidden centers from the training set and a detailed selection procedure are presented. The solution to orthogonal decomposition terms under the condition of varying hidden centers is obtained theoretically. Finally, the effectiveness and efficiency of our proposed approach are demonstrated by two examples.