Released 2011
CoSyNE is a neuroevolution method where synapses of the network are evolved in separate subpopulations in a cooperative fashion. It is particularly powerful in control tasks such as double pole balancing, where it has performed better than other methods. The code is structured so that it is easy to apply it to other tasks by changing the environment class.
Download:
TAR
Faustino Gomez Postdoctoral Alumni tino [at] idsia ch
Accelerated Neural Evolution through Cooperatively Coevolved Synapses 2008
Faustino Gomez, Juergen Schmidhuber, and Risto Miikkulainen, Journal of Machine Learning Research (2008), pp. 937-965.
Efficient Non-Linear Control through Neuroevolution 2006
Faustino Gomez, Juergen Schmidhuber, and Risto Miikkulainen, In Proceedings of the European Conference on Machine Learning, pp. 654-662, Berlin 2006. Springer.