A neural computation for canonical representations of nonlinear functions
Qing Huang, R.-W. Liu · 2002
An efficient constructive method to compute a canonical representation of any continuous nonlinear function with a neural network is presented. This neural network consists of only one hidden layer and the number of nodes in the hidden layer is estimated. A particular initial condition is chosen so that the backpropagation algorithm converges to a solution with an error between the neural network realization and the given function less than a given delta >0. Two simulation examples are used to demonstrate the method, and a simple application of this method to the design of a negative device is given.>