A parameter-separable learning algorithm for multilayer feedforward networks

Summary form only given. The higher the nonlinearity of a system to be identified, the larger is the structure of network as well as the quantity of parameters needed. Algorithms which adjust all parameters simultaneously extend the learning process. The parameters can be classified into linear and nonlinear ones after analysis on several typical networks (BP, RBF and fuzzy neural network), and based on this the paper carries out a parameter-separable learning algorithm to accelerate the learning process.