Uncovering the Principles of Brain-Computer Interface Learning

Brain-Computer Interfaces (BCIs) read neural activity to infer the user’s intent and use it to control effectors, enabling communication through thought. Therefore, BCIs offer a unique opportunity to restore function to patients with movement and even cognitive/affective conditions. Despite growing interest and investment in BCIs, most efforts are directed towards improving decoder algorithms, while little is understood about how the brain learns to control BCIs. Does learning occur locally, via plasticity in the targeted area, or globally, through interactions across a multi-region network? Here, I address this question by reading neural activity from the striatum. Neuropixels probes allow recording from both the striatum and multiple upstream cortical areas during BCI learning. First, I will build a biologically plausible, data-constrained model of the corticostriatal circuit by fitting neural activity (electrophysiology and dopamine photometry) and examine different mechanisms of BCI learning. Then, I will evaluate the model’s predictions in a series of novel mouse BCI experiments, specifically designed to distinguish between local and global BCI learning. Overall, this hypothesis-driven approach will not only elucidate the neural mechanisms of BCI learning, but will also inspire innovation in BCI design, informing next-generation clinical applications.