UC San Diego

Key advancements in recording hardware, data computation, clinical care, and cognitive science continue to drive new possibilities in how humans and machines can interact directly through thought. Neural data analyses with these advancements has progressed neuroscience research in functional brain mapping and brain-computer interfaces (BCIs). Much of our knowledge about BCIs is informed by data collected through carefully controlled experiments. Constraining BCI experiments with structured paradigms allows researchers to collect a high number of consistent data in a short amount of time, while also controlling for external confounds. Very little is currently known about how well these task-based relationships extend to daily life, in part because collecting data outside of the lab is challenging. To further understand natural brain activity, we must study more complex behaviors in more environmentally relevant settings. The results of this dissertation address three general challenges to studying neural correlates to unstructured behaviors. First, we continuously monitored unstructured human movements in the epilepsy monitoring unit using a video sensor synchronized to clinical intracortical electrodes. Second, we annotated unstructured behaviors from these video using both manual and computer vision methods. Finally, analyzed neural features with respect to unstructured human movements, and evaluated the performance of features identified in previous task-based studies. The preliminary nature of this work means that a majority of our demonstrations are whether the continuous paradigm can be leveraged, how one might go about leveraging it, and evaluations that tie our results back to earlier task-based studies. Our advances here motivate future works that focus more intently on what types of behaviors and neural signal features to explore.