The development of neuroprosthetic devices promises to allow
previously immobile patients to control the movement of an external
device using only their brains' electrical activity. Prediction
algorithms used to control such devices rely on models relating the
firing rates of a population of neurons to intended movement variables
such as direction. However, since no data on real arm movement will
be available prior to use of the prosthetic, and recent research has
shown that neurons may change their firing patterns in response to
visual feedback, an algorithm is needed that in addition to predicting
movement can also perform real time estimation of the model. This
article proposes statistical methods for performing these related
tasks and demonstrates the methods' effectiveness using data taken
from a series of experiments using a rhesus monkey.
Keywords: neuroprosthetics, state-space models, particle filtering,
adaptive methods