Con't Neural Nets
Volume Number: 6
Issue Number: 11
Column Tag: Programmer's Workshop
Continuous Neural Networks 
By Wayne Joerding, Pullman, WA
Code for Continuous Neural Networks
For the past few years there has been considerable renewed interest in the use of
Neural Networks in such diverse areas as cognitive science, artificial intelligence, and
physics. My interest in the subject derives from a paper by K. Hornik, M.
Stinchcombe, and H. White who show that a three layer feedforward neural network
can approximate any continuous nonlinear function to any desired degree of accuracy
by expanding the number of units in the hidden layer. This result probably explains
the success physicist have had in using neural networks to model nonlinear dynamic
systems, i.e. chaos. I plan to investigate the use of neural networks to model economic
behavior (yes, I’m an economist).
Programming neural networks presents several problems, among them the
algorithms which compute an output for the network and a method of training the