Abstract:
In artificial neural networks, the parameters
may include the number of layers, the number of
hidden units, the activation function and the algorithm
parameters such as learning rate for optimization.
Many researchers have proven that the training of
artificial neural networks is a complex process and
methods of training are highly varied. Some attempt
to approximate the process of biological neurons but
many diverge greatly from them in an attempt to find
more computationally efficient methods to achieve
optimal or near-optimal weights. Although radialbasis function networks (RBF) are well known for
requiring short training period among artificial
neural networks, these methods perform a local
search and they can easily fall in local minima by
producing sub-optimal solutions. Therefore, the
performance of network training is not good and the
accuracy is low for RBF neural networks. The
traditional network weight training generally uses
gradient descent method and it can not get the global
optimum. Training the weights by optimization
method can find the weight set that approaches global
optimum while do not need to compute gradient
information and it can help to reduce error rate in
network training .Clonal selection algorithm is a
global search among optimization method and it can
provide an efficient alternative for the optimization of
neural networks. In this paper, we use clonal selection
algorithm to adjust weight units which are important
to improve network training in RBF neural network.
