Learning with integer weighted functional link preceptron

In this paper, an algorithm for the design of functional link single layer neural networks using N binary inputs is described. The resulting connectio n weights are pure integer numbers. These integer weights facilitate faster learning by the neural network due to binary operations rather than algebr aic multiplications. Comparison between various learning algorithms, mainly backpropagation with the delta rule, and the functional link approach is d emonstrated via several recognition applications. It is illustrated that th e functional link approach, due to enhancing input patterns, produces a rob ust algorithm for linearly non-separable classification problems in terms o f processing speed and convergence. Furthermore, implementation of the neur al network can be accomplished using the vast availability of off-the-shelf components and VLSI techniques.