Uma ferramenta de visualização para redes neurais artificiais do tipo neocognitron

Abstract

Some Artificial Neural Networks may increase their size significantly with its use, making it difficult for a human being study it well. The networks of the Neocognitron model are an example of this kind of network, increasing its initial size many times before the end of their training. Using Scientific Visualization techniques, we obtained a tool that allows the user to see in details a neocognitron network actually working, be it during its training time or the recognition process. This allows a better comprehension of the whole model , as it helps in the construction of a network by allowing experimentation with the parameters of the model. Powering this tool is a library of classes, that can be used separately to the implementation of new ways of visualization or in new simulators of other models of neural networks.