Skip to main content
. 2025 Jan 20;31(1):96–104. doi: 10.1038/s41591-024-03341-8

Fig. 2. Dimensionality.

Fig. 2

a, Dimensionality of the neural activity during closed-loop decoding using either the 2D decoder/task or the 4D decoder/task for either one or two new finger-group targets (1 FG or 2 FG) per trial. Light blue dots represent data from a day, and black diamonds represent the mean, µ. b, Summary statistics comparing the 2D decoder on the 2D task (2D, 2T; n = 233 trials), the 4D decoder on the 2D task (4D, 2T; n = 329 trials) and the 4D decoder on the 4D task (4D, 4T; n = 284 trials) based on the acquisition time, time to target, orbiting time, acquisition rate, path length efficiency and the percent of trials successfully completed. The error bars represent the standard error of the mean. c, A typical online block showing the decoded index–middle finger group velocities using the 2D on 2T (blue) during an online block. Offline, the 4D decoding algorithm was used to predict index–middle group velocities from the same block (orange). The normalized CC between the online and offline signals is given in the bottom-right corner. Units are denoted so that the range of motion for each DOF is unity. d, For the ten blocks on the 2D task, the 4D decoding algorithm was used to predict finger velocities during online blocks using the 2D decoder (online 2D in blue, offline 4D in orange), and the 2D decoding algorithm was used to predict online velocities using the 4D decoder (online 4D in blue, offline 2D in orange). CC between the offline and online signals is represented by dots and averaged across both finger groups. The diamond and µ denote the mean value for the average of all five blocks for each paired comparison.