RBF neural network is widely used [1–3] in the traditional classification
problem. Comparing the RBF neural network with the classic forward neural network such as back-propagation (BP) network [4], the main difference is that BRF neural purchase BX-912 network has more hidden layer neurons, only one set of layer connection weights from the hidden layer to the output layer; the hidden layer takes the radial basis function as the activation function, generally using Gaussian function [5]; both unsupervised and supervised learning have been used in the training process and so on. In the hidden layer of RBF neural network, each neuron corresponds to a vector of the same length as a single sample, which is the center of neuron. The centers are usually
obtained by K-means clustering; this step seems as unsupervised learning; the connection weights from the hidden layer to the output layer are usually obtained by the least mean square (LMS) method, so this step seems as supervised learning. In the RBF neural network, the nonlinear transfer functions (i.e., basis function) do not affect the neural network performance very much; the key is the selection of the center vectors of basis functions (hereinafter referred to as the “center”). If we select improper center, it is difficult for the RBF neural network performance to achieve satisfactory results; for example, if some centers are too close, they will produce approximate linear correlation and then result in lesions on numerical criteria; if some centers are too far, they are short of the requirement of linear processing. Too many centers may easily lead to overfitting, while it is difficult to complete classification tasks if centers are too few [6]. RBF neural network performance
depends on the choice of the hidden layer’s center, it determines whether the neural network had successful training and can be applied in practice or not. Genetic algorithm (GA) is developed from natural selection and evolutionary mechanisms; it is a search algorithm with the characters of being highly parallel, randomized, and adaptive. Genetic algorithm uses the group search technology and takes population on behalf of the solution of a group questions. By doing a series of genetic operations like selection, crossover, mutation, and so on to produce the new generation population, Drug_discovery and gradually evolve until getting the optimal state with approximate optimal solution, the integration of the genetic algorithm and neural network algorithm had achieved great success and was widespread [7–10]. Using the genetic algorithms to optimize the RBF neural network is mostly single optimizing the connection weights or network structure, [11–13], so in order to get the best effect of RBF, in this paper, the way of evolving both two aspects simultaneously is provided.