We evolve small continuous-time recurrent networks with fixed weights that perform Hebbian learning behavior. We describe the performance of the best and smallest successful system, providing an in-depth analysis of its evolved mechanisms. Learning is shown to arise from the interaction between the multiple timescale dynamics. In particular, we show how the fast-time dynamics alter the slow-time dynamics, which in turn shapes the local behavior around the equilibrium points of the fast components by acting as a parameter to them.
Izquierdo-Torres, E. and Harvey, I. (2007): Hebbian Learning using Fixed Weight Evolved Dynamical ‘Neural’ Networks. In H. A. Abbass et al. (eds.), Proceedings of the First IEEE Symposium on Artificial Life. (IEEE-ALife’07). Pp. 394-401. ISBN: 1-4244-0698-6. IEEE Press.