Sensorimotor Perturbations to Study Neural Computations of Motor Learning

Abstract

Motor learning depends on the brain’s ability to link actions to their outcomes and learn/refine sensorimotor maps through feedback. This thesis examines how sensorimotor perturbations using Brain Computer Interfaces (BCIs) and reaching behavioral tasks paradigms reveal the neural computations that support learning. We investigated BCI learning experiments with assistive manipulations that improve performance and found that assistive manipulations alter credit assignment learning and drive compaction of learned neural representations concentrating task information in fewer neurons. Second, we adapted visuomotor perturbations, commonly used in motor learning studies, to a new interface that preserves a portion of natural redundancy to study how task relevance and altering task relevance influenced learning. These perturbation paradigms revealed distinct signatures of error based and new controller learning when task relevance is manipulated in redundant spaces. Next, we introduce an electrocorticography based brain computer interface combined with optogenetic stimulation to investigate how rest and network connectivity contribute to learning that occurs outside active practice. Finally, we compared linear and nonlinear approaches for capturing the 'dimensionality'- complexity of activity patterns across neural populations.Together, these studies present a set of complementary tools that each uncover a different aspect of learning and offer findings that can shape future learning experiments and brain computer interface design. Future studies can integrate these tools, including redundancy-based perturbation paradigms, assistive interventions, measures of network connectivity, and analyses of population structure. Combining these approaches may enable frameworks that link neural computation with large scale network organization and deepen our understanding of the principles that support flexible learning in the brain.