A self-organizing reasoning neural network

In this paper, the design of the feedforward neural network is formulated as a Bayesian-Gaussian reasoning problem. The output of the network is an optimal solution of the a posteriori probability density function. The parameters of the network can be set immediately when the training data are known. The self-organizing ability of the network can also be easily designed based on the reasoning model, thus even when the training samples become increasingly large, the network is able to optimize its parameters in an optimal way. Simulation shows that the reasoning neural network is comparable to or better than the well-known BP network in nonlinear stochastic prediction problems, and the self-organizing neural network is superior to the fixed parameter network in performance