The adoption of structures based on cellular automata for the spatial organisation of populations in evolutionary algorithms has been shown to improve performances of the synthetised models. In this work we review recent research on the integration of cellular structures into geometric semantic genetic programming, including variants that aim at producing more interpretable models. In the setting of symbolic regression on real world data, the research findings provide evidence to the previous claims, as well as give insights about the interplay between the studied approaches, the accuracy and the size of the resulting models.
