Evolution-based optimization of agents driven by the free-energy principle
In this thesis the behavior and tness of free energy agents subject to evolutionary principles is studied. The free energy parameters of these agents are optimized by an evolutionary algorithm, of which the e ciency in the process of parameter optimization is analyzed. The existing research on the free energy principle has already been focusing on how accurate and useful the free energy model might be in the context of real life organisms. However, the research has mostly only hinted at the possibilities of evolu- tionary integration. In this thesis, a model integrating both the free energy principle as well as evolutionary computing is introduced, which simulates the evolutionary process of a population of 200 agents. It has been found that through the optimization of the free energy parameters, the tness of at least 90% of all agents is steadily improving throughout the evolutionary process, with the steepest increase happening in the rst 25 generations. The evolutionary algorithm was able to transform a completely randomized population into a population of which at least 80% of all agents were t for survival. This thesis was written in the hopes of further closing the gap between neurology and mathematics, as well as contributing to the evidence behind the biological validity of the free energy principle.
Faculteit der Sociale Wetenschappen