Fig. 8: Results of an example RNN model.

a Schematic of the RNN model. The RNN model consisted of an input layer, a hidden layer, and an output layer. The input layer received a signal for the position of two targets simultaneously, while the output layer produced a PV, whose magnitude reflects the degree of movement tendency for the neural population in the corresponding direction. b Response of two example nodes under four conditions. The selected conditions are represented in different colors, as shown on the left. The black dots denote the target on (TO), the first movement onset (MO), and the second movement onset (MO2), respectively. c Full model fitting result for RNN nodes. It turned out that the temporal patterns of the coefficients in this RNN model are comparable to those in real neurons. The error band in this panel was plotted in mean ± 2 standard error. d The PCA neural states indicate initial states, during motor preparation. Colors indicate the first movement directions; DR trials are presented in the same color family as related SR trials. Markers indicate the second reaching direction. e Quantification of the clustering in (d). The normalized Euclidean distance of states under different DR conditions was calculated against the state under 180° condition, in the first reach-direction groups; the normalized Euclidean distance of states under different first reach conditions was calculated against the state under 0° condition, in the DR-condition groups.