Fig. 4: Real-time feedback task.
From: Voluntary control of semantic neural representations by imagery with conflicting visual stimulation
a Timing of ECoG acquisition and feedback image presentation during the real-time feedback task is shown with a schematic. Immediately after delivery of the instruction, 1000-ms ECoGs were acquired while the subject was watching the black screen to infer the semantic vector in real-time (online vector: vonline) with the online decoder. Next, the first feedback image was selected from among the images in the training videos based on the highest \(R({v}_{online},\,{v}_{true}^{i})\) and presented to the subject. Then, successive 1000-ms ECoGs were acquired every 250 ms to calculate the next online vector (vonline), which was the linear interpolation between the inferred vector from the ECoGs and the previous online vector. A total of 32 feedback images were presented for each instruction. The order of instructions was randomized. The system delay from the end of the acquisition to the image presentation was approximately 200 ms. All four subjects participated in four sessions of 30 trials each, with breaks between sessions to minimize their fatigue. b Locations of subdural electrodes used for decoding in the real-time feedback task are mapped on a normalized brain surface. Red and blue markers denote electrodes in the left and right hemispheres, respectively. The number of depth electrodes used in the real-time decoding is shown in parentheses.
