Induction of neural networks for parallel binary operations

The author achieves reproducibility of synaptic weights by using a neural network architecture called the Classitron. It is possible to induce parallel algorithms for the general case by training smaller networks. This is shown by producing a parallel carry-less addition scheme of n binary numbers, each m bits long. A particular advantage of the Classitron is the specification of internal representation via nonlinear functionalities which can be translated easily to the number of hidden nodes of a multilayer perceptron network