L'algorisme de retropropagació d'errors des d'un punt de vista matemàtic

Abstract

[en] In this project, we are going to study the backpropagation algorithm from a mathematical perspective and analize why this algorithm, which was forgotten in the past, is now used in supervised learning in order to train artificial neural networks. First of all we will talk about how artificial neural networks are organized and how they work. Next, we will study the gradient descend method, the Newton’s method, and other methods, used in learning. Finally, we will see how the backpropagation algorithm works and the calculations derived from it.