Investigating the neural basis of learning using brain-computer interfaces

Learning a new skill requires one to produce new patterns of activity among networks of neurons. This applies not only to physical skills, such as learning to play a new sport, but also to abstract skills, such as learning to play chess. An abstract skill that we can use to study the neural mechanisms of learning, in general, is controlling a brain-computer interface (BCI). BCIs were conceived of as assistive devices to help people with paralysis, limb-loss, or other neurological disorder, but they have also proven effective as tools to study the neural basis of sensory-motor control and learning. We tested the ability of subjects to generate neural activity patterns required to control arbitrary BCI decoders. We found that the subjects could more easily learn to control the decoder when they could use existing patterns of neural activity than when they needed to generate new patterns. We also analyzed the way in which subjects adapted their neural activity during learning. We found that neural activity adapts in a way that is consistent with the learning-related performance improvements and that the trial-to-trial variability of neural activity decreased as performance improved. We tested how specific properties of BCI decoders, which translate neural activity into movements of the effector, influence the ability to learn to control a BCI by incorporating dimensionality reduction into a Kalman filter and assessing how performance related to the number of latent dimensions. We found that the subjects could use a standard Kalman filter just as well as a Kalman filter that incorporates dimensionality reduction. However, as the dimensionality of the model increased, performance improved up to an asymptotic level. Lastly, we tested whether increasing the difficulty of a task would lead the subjects to learn to demonstrate better BCI performance. We implemented an instructed path task that required the animals to move a cursor along re-defined paths, and we found that this task motivated one monkey to improve his performance. In all, these studies help to uncover what contributes to BCI control, and they help pave the way for transitioning BCIs from the lab to the clinic