A self-optimizing stochastic dynamic node learning algorithm for layered neural networks

Abstract

Summary form only given, as follows. The random optimization method typically uses a Gaussian probability density function (PDF) to generate a random search vector. In the present work the random search technique was applied to the neural network training problem and was modified to dynamically seek out the optimal probability density function (OPDF) from which to select the search vector. The dynamic OPDF search process, an autoadaptive stratified sampling technique, and a dynamic node architecture (DNA) learning scheme complete the modifications of the basic method. The DNA technique determines the appropriate number of hidden nodes needed for a given training problem. Using DNA, the neural network architectures do not have to be set before training is initiated. The approach was applied to networks of generalized, fully interconnected continuous perceptrons. Computer simulation results were obtained. >