Evolving developing spiking neural networks

Indirect encoding strategies aim at higher evolvability by reducing the dimensionality of the search space. If on one hand scalability is often improved for specific tasks, on the other the generality of these methods can be limited. In previous work, a development system was introduced and tested in the evolution of specific 2D morphologies of various size and complexity. Here the same model is used to instead specify the structure and properties of neuro-controllers for simulated Khepera robots. In this paper, we introduce a plastic spiking neural network model, particularly suited for evolution and development, testing its performance against direct encoding. Compared to previous work, the new task implies the solution of a functional problem. Nevertheless, results show similar conclusions regarding the improved scalability of the development system and its connection to regularity