Using Evolutionary Methods to Parameterize Neural Models: a Study of the Lamprey Central Pattern Generator

The neural controller of anguilliform swimming in lampreys is particularly well studied because of its relative robustness and simplicity. In this chapter we look at connectionist models of the controller – in which populations of similar neurons are represented by abstract, differential-equation-based models – and describe the use of evolutionary computation techniques for investigating the space of appropriate architectures and parameters for such models. An introduction to this style of modeling is followed by a presentation of the lamprey central pattern generator model, devised by Ekeberg, and its development by Ijspeert using genetic algorithms. Some results on the robustness to body variation in the modelled controllers will be described, and the value of this approach to neural modeling work will be discussed.