Coevolution of functional flow processing networks

We present a study about the construction of functional flow processing networks that produce prescribed output patterns (target functions). The constructions are performed with a process of mutations and selections by an annealing-like algorithm. We consider the coevolution of the prescribed target functions during the optimization processes. We propose three different paths for these coevolutions in order to evolve from a simple initial function to a more complex final one. We compute several network properties during the optimizations by using the different path-coevolutions as mean values over network ensembles. As a function of the number of iterations of the optimization we find a similar behavior like a phase transition in the network structures. This result can be seen clearly in the mean motif distributions of the constructed networks. Coevolution allows to identify that feed-forward loops are responsible for the development of the temporal response of these systems. Finally, we observe that with a large number of iterations the optimized networks present similar properties despite the path-coevolution we employed.