Neural network construction and rapid learning for system identification

A new learning algorithm is introduced and used to identify nonlinear functions in a feedforward neural network. Its distinctive features are that it transforms the problem to a quadratic optimization problem that is solved by a number of linear equations and it constructs the appropriate network that will meet the specifications. The quadratic optimization/dependence identification algorithm extends the results of quadratic optimization single layer network training and significantly speeds up learning in feedforward multilayer neural networks compared to standard backpropagation.<
>