SUMMARY
One hallmark of natural motor control is the brain’s ability to adapt to perturbations ranging from temporary visual-motor rotations to paresis caused by stroke. These adaptations require modifications of learned neural patterns that can span the time-course of minutes to months. Previous work has shown that populations of neurons fire on coordinated low-dimensional subspaces that are resistant to changes, and perturbations requiring neural activity to move outside of these subspaces are difficult to learn. Subsequently, perturbations that remain within the neural subspace are easier to adapt to. However, it is unclear how motor cortex might respond to perturbations whilst still learning. To answer this question, five nonhuman primates were used in three brain-machine interface (BMI) experiments, which allowed us to track how specific populations of neurons changed firing patterns as task performance improved. In each experiment, neural intentions were estimated with biomimetic decoders that were periodically refit, creating perturbations throughout learning. We found that decoder perturbations caused neurons to increase exploratory patterns on within-day timescales without hindering previously consolidated patterns regardless of task performance. The flexible modulation of these exploratory patterns in contrast to relatively stable consolidated activity suggests a concomitant exploration-exploitation strategy that adapts existing neural patterns during learning.