New neuroinspired algorithms in Artificial Neuro-Astrocytic Networks

Abstract

This thesis presents a series of complex algorithms based on classical Artificial Neuron Networks (ANN) designed from a multidisciplinary perspective that brings together the fields of Artificial Intelligence and Neuroscience, seeking to represent in silico the tripartite synapse. All algorithms are designed as a complex circuit in which an ANN modulates a main ANN to improve its performance. The central algorithm in this thesis is the so-called DAM, which relies on an artificial astrocyte that also modulates the circuit of two RNAs mentioned above. To study the influence of an artificial astrocyte on the circuit, two other control algorithms are designed. The first one, DM, is composed exclusively of the second ANN modulating the main ANN; while the second one, CONTROL, is an algorithm that as an intermediate modulating element has an element that has an internal functioning identical to that of the other artificial neurons that compose the rest of the circuit. A total of four behaviors of the artificial astrocyte inspired by different biological realities were designed to explore if there were any of the chosen ones that most favored our algorithm when solving Machine Learning problems.