RNS neural networks
Giovanni Martinelli, R. Perfetti · 2002
A type of perceptron based on a residue number system (RNS) implementation is introduced. Due to the scaling involved in the RNS, it operates with a particular periodic nonlinearity which increases its capability of problem solving. Two examples which clearly show that the RNS perceptron has better performance than the classical perceptron are presented. In both the examples the RNS perceptron solves a problem which is out of the possibility of the corresponding nonRNS perceptron.>