BAND: Behavior-Aligned Neural Dynamics is all you need to capture motor corrections
Motor cortical neural activity is commonly viewed as a low-dimensional dynamics evolving from the movement preparation state, which explains the most of both neural and behavioral variability. We found that movement corrections to unexpected behavior perturbations do not follow the same pattern, with only a small fraction of neural variability explaining large changes in behavior. We show that capturing both movement planning and corrections requires models that incorporate dynamics and weak behavior supervision. We characterize the bidirectional relationships between motor cortical activity and behavior, identifying neural code for both feedforward and feedback-driven motor control.
Jan 1, 2025
