Estructura de redes neuronales (MLP) y su aplicación como aproximador universal

Abstract

The present work is oriented to the study of the theoretical support of artificial neural networks (MPL - Multilayer Perceptron), compared to their classification capacity. For this purpose, the concepts that allow their mathematical understanding are formally defined, in addition to the implementation of said notions in the elaboration of a code that shows, through particular examples, the classification capacity of neural networks based on the theory developed here. Additionally, the ability of neural networks to approximate functions is evidenced, through an example applied to the developed code, this result is possible thanks to the Universal Approximation Theorem of Functions.