Table 2.

Decoding accuracy for position and velocity across five sessions, with comparison to previous work. The table shows $R^2$ accuracy on test data, with position decoding in white cells and velocity decoding in gray cells. Dataset size and neuronal population are listed for each session, with the bolded row representing the example session used in Figs. 1-3 and 5-8. We also include decoding performance from Glaser et al.,²⁷ who decoded the velocity of a cursor controlled by a manipulandum from motor cortex recordings, as a comparison between skeletomotor and oculomotor decoding performance, since they used the same decoders and developed the coding package we used.

Dataset	Dataset size (min)	Population size (neurons)	Decoding accuracy ($R^2$)
			wf	wc	xgb	svr	dnn	rnn	gru	lstm
Session 26	19.3	73	0.85	0.86	0.85	0.90	0.91	0.89	0.93	0.93
			0.51	0.50	0.49	0.54	0.54	0.55	0.57	0.57
Session 29	63.6	65	0.72	0.72	0.73	0.80	0.88	0.87	0.89	0.89
			0.50	0.50	0.48	0.52	0.60	0.62	0.64	0.64
Session 24	60.1	52	0.56	0.56	0.60	0.68	0.81	0.78	0.81	0.81
			0.46	0.45	0.49	0.53	0.61	0.63	0.66	0.66
Session 22	46.6	52	0.40	0.40	0.42	0.52	0.65	0.65	0.66	0.66
			0.36	0.34	0.35	0.42	0.48	0.51	0.52	0.52
Session 32	47.6	55	0.42	0.43	0.40	0.49	0.60	0.59	0.59	0.62
			0.31	0.32	0.28	0.35	0.42	0.45	0.45	0.47
Glaser et al.27	10	164	0.78	0.79	0.82	0.81	0.86	0.84	0.87	0.88