Um Controlador da Din�mica da Aritm�tica de M�ltiplas Colunas usando Redes Neurais Artificiais

Abstract

This work d escribes a series of improvements on Dallaway´s neural network-based model for multiple column arithmetics. Both the original model and the improvements herein reported are firmly rooted in data collected in studies on how children are actually taught and learn arithmetics. Several changes in the training data, the training regime and the actions that control the execution of the arithmetic operations-which are the neural network´s outputs-were carried on. This way a more adequate model was developed, in terms of the generalization ability of the network that underlies the model, the correction of the results produced and, most important, the cognitive plausibility of the errors generated by the model.