Warming up recurrent neural networks to maximise reachable multistability greatly improves learning

Open in new window