Neural network as an alternative computational paradigm

Abstract

A great number of research activities in neural networks have been concerned on its powerfulability to mimic human "qualitative reasoning" especially applied to pattern recognition problems. Less work has been performed in applying neural network to process floating point numbers due to its seemingly inherent inaccuracy. However, the parallel nature of neural network has a potential for high speed massive processing of data making neural network "number crunching." This ability to process floating point numbers in a massively parallel fashion makes neural network a belter alternative to sequential machines in handling enormous volume of data. Furthermore, with its ability to generalize a mapping rule from a finite number of data in a given training set, it offers an alternative approach to solving problems via learning, in which solution to a problem is learned from a set of examples as compared to the conventional method which is based on an explicit set of programmed instructions.