Abstract
We study, on the basis of a simple neural–network like system that we call a spatial network, how several properties of two evolvable systems: a nervous system and computer software, could influence on the evolvability of a genetic algorithm. The following properties are modelled: hierarchical modularity, mirroring, extensive variability and quantitative genetic traits. By mirroring we understand mimicking of the structure of processed data by the structure of the processing algorithm. We estimate the evolvability of several examples of the presented networks, which perform simple image processing tasks, like stereo–to–depth. The evaluation is done by studying the size of these networks' genotype.