Evolutionary Optimization of Echo State Networks: Multiple Motor Pattern Learning

Echo State Networks are a special class of recurrent neural networks, that are well suited for attractor based learning of motor patterns. Using structural multiobjective optimization, the tradeoff between network size and accuracy can be identified. This allows to choose a feasible model capacity for a follow-up full weight optimization. Both optimization steps can be combined into a nested, hierarchical optimization procedure. It is shown to produce small and efficient networks, that are capable of storing multiple motor patterns in a single net. Especially the smaller networks can interpolate between learned patterns using bifurcation inputs.