Fig. 5: Demonstration of NoMAD and retrospective target inference (RTI) on closed-loop iBCI data.
From: Stabilizing brain-computer interfaces through alignment of latent dynamics
a Schematic of the center-out cursor control task, modified from figure in Willett et al. 6. On any given trial, the participant controlled the cursor from the center to one of the outer four targets. Upon successful target acquisition, the cursor was automatically returned to the center. b, c R2 of predicted cursor velocity compared to estimated intended cursor velocity for each of the methods during two sessions. Each initial calibration block’s decoding performance is denoted with an x. The first evaluation block, during which no recalibration has yet taken place, has decoding performance denoted by an open circle marker. Decoding performance after recalibration and/or alignment has been performed using all previous blocks is denoted using closed circle markers.
