The concepts of muscle activity generation driven by upper limb kinematics

Table 2 Overview of the performance for all architectures (recurrent neural network (RNN), sub-sequenced input recurrent neural network (RNNseq), feedforward neural network (FNN), sub-sequenced input feedforward neural network (FNNseq), convolutional neural network (CNN)) and channels (electrodes 1-8) with the general approach (training and testing on multiple subjects) evaluated by the zero-line score Z_s, mean square error MSE, the squared correlation coefficient r², and the coefficient of determination R²

		1	2	3	4	5	6	7	8	Average
RNN	Z_s	89.53	91.89	90.31	88.75	81.27	89.45	80.55	80.44	88.13
	MSE	0.0009	0.00131	0.00285	0.00203	0.00269	0.00066	0.00066	0.00081	0.00149
	r²	0.88156	0.89965	0.87858	0.84992	0.77071	0.87266	0.79592	0.80298	0.85597
	R²	0.87719	0.89916	0.87292	0.8474	0.75624	0.8695	0.77816	0.78616	0.84904
RNNseq	Z_s	86.24	90.34	86.78	84.71	69.16	84.42	76.11	72.94	83.33
RNNseq	MSE	0.00119	0.00156	0.00388	0.00275	0.00443	0.00098	0.00081	0.00113	0.00209
FNN	Z_s	88.39	90.97	90.06	88.43	63.58	88.59	79.81	82.19	85.21
FNN	MSE	0.001	0.00146	0.00292	0.00208	0.00523	0.00071	0.00068	0.00074	0.00185
FNNseq	Z_s	88.6	91.67	89.74	88.57	75.35	86.73	78.21	81.71	86.81
FNNseq	MSE	0.00102	0.00139	0.00311	0.00214	0.00356	0.00085	0.00077	0.00078	0.0017
CNN	Z_s	88.49	91.24	89.78	88.25	71.94	86.59	79.89	81.0	86.18
CNN	MSE	0.00099	0.00141	0.003	0.00212	0.00403	0.00084	0.00068	0.00079	0.00173

ISSN: 1475-925X