The structure optimization of radial basis probabilistic neural networks based on genetic algorithms

In this paper, the genetic algorithm is introduced into the structure optimization of radial basis probabilistic neural networks. A special encoding method of individuals is proposed in this paper. The encoding method involves not only the number but also the locations of selected hidden centers. In addition, for this encoding method we construct a fitness function for precision control. To speed up the training we employ the method of matrix pseudo-inverse to train the corresponding networks. Finally, we take the telling-two-spirals-apart problem, for example, to validate the ability of the genetic algorithm for the structure optimization of radial basis probabilistic neural networks