Fig. 4: NoMAD enables stable offline decoding across 5 weeks in a reaching task.

a Schematic of the center-out reaching task, modified from figures in Ethier et al.⁶⁸ and Perich et al.⁴⁰ Copyright (2018), with permission from Elsevier. b R² of aligned cursor velocity decoding for each method applied to all pairs of sessions (12 sessions spanning 38 days). c Median R² of cursor velocity within each 5-day bin. Conventions as in Fig. 3. d Top: Percentage of Day K decoding failures (R² < 0) in 5-day bins. Bottom: Decoding performance as a function of days between sessions. Conventions as in Fig. 3. e Left: Measured single-trial reach trajectories, colored by target location. Right: reach position trajectories integrated from the decoded reach velocity when Day 38 is aligned to Day 0.