Modelos de procesamiento de la información en el cerebro aplicados a Sistemas ConexionistasRedes NeuroGliales Artificiales y Deep Learning

Abstract

In the field of Artificial Intelligence, connectionist systems have been inspired by neurons and, according to the classical view of neuroscience, they were the only cells capable of processing information. The latest advances in Neuroscience have shown that glial cells have a key role in the processing of information in the brain. Based on these discoveries, Artificial NeuroGlial Networks (RNGA) have been developed, which have two types of processing elements, neurons and astrocytes. In this thesis, this line of multidisciplinary research that combines Neuroscience and Artificial Intelligence has been continued. For this goal, a new behavior of the astrocytes that act on the output of the neurons in the RNGA has been developed. A comparison has been made with the Artificial Neuron Networks (ANN) in five classification problems and it has been demonstrated that the new behavior of the astrocytes significantly improves the results. After prove the capacity of astrocytes for information processing, in this thesis has been developed a new methodology that allows for the first time the creation of Deep Learning networks containing thousands of neurons and astrocytes, called Deep Neuron-Astrocyte Networks (DANAN). After testing them in a regression problem, the DANAN obtain better results than ANN. This allows testing more complexes astrocyte behaviors in Deep Learning networks, and even creates astrocyte networks in the near future.