A constructive neural network algorithm for function approximation
Timothy J. Draelos, Donald R. Hush · 2002
A study of the approximation capabilities of single hidden layer neural networks leads to a strong motivation for investigating constructive learning techniques as a means of realizing established error bounds. Learning characteristics employed by constructive algorithms provide ideas for development of new algorithms applicable to the function approximation problem. A novel constructive algorithm, the iterative incremental function approximation (IIFA) algorithm is presented in detail. The algorithm operates in polynomial time and is demonstrated on one and two dimensional function approximation problems.