Additional perspectives on feedforward neural-nets and the functional-link

It has been proved that multilayer feedforward neural-nets with as few as a single hidden layer can serve as universal approximators of functions mapping multidimensional space R^s to one-dimensional space R. Our prior experience has provided us with ample pragmatic evidence that the model can be simplified, with use of functional-links which need not be learned. In this paper, we prove theorems which provide a theoretical justification for use of the highly efficient functional-link approach.