A Novel Method for Training an Echo State Network with Feedback-Error Learning

Table 2

The mean NRMSE and variance for 20 repetitions of each experiment. All the networks were tested on one repetition of the YMCA movement.

Experiment	Position error	Var	Motor error	Var
1 Orig. method, 1 rep.	0.4000	0.0024	0.4737	1.4𝐸−04
2 Orig. method, 5 rep.	0.1088	0.0125	0.2435	0.0071
3 Orig. method, 10 rep.	0.1193	0.0081	0.2500	0.0066
4 New method, 𝛽=0.3	0.0494	0.0018	0.1100	6.9𝐸−04
5 New method, 𝛽=0.1	0.0245	7.0𝐸−05	0.0717	1.4𝐸−04
6 New method, 𝛽=0.05	0.0385	0.0020	0.0669	9.2𝐸−04