Agentes de software basados en técnicas de aprendizaje automático. Perspectivas desde 2010 hasta 2023

Abstract

This study aims to analyze the main theoretical and practical proposals in which software agents have been integrated with machine learning models to determine their scope in terms of intelligence, proactivity, collaboration and learning. For the development of this research, the methodology proposed by Kofod-Peterson was carried out. Applying the methodology, 55 studies were analyzed. The studies showed that in the interaction between software agents and machine learning, cooperative and collaborative processes have been widely used in the resolution of control problems and in the optimization of data in distributed scenarios such as home, games and telecommunication. It was also found that mostly reinforcement learning models were used compared to machine learning models because they contribute more significantly to cooperative task modeling, which is widely used in intelligent systems.