An investigation into time structure of EEG-based cursor control in a brain-computer interface using machine-classification methods

Brain-computer interface (BCI) research combines neuroscience, motor learning and computer science and is therefore very interesting for AI researchers. A BCI can help when the natural interface between human and computer, i.e. motor control over the arms, is deficient as in quadriplegics. A BCI uses signals from the brain to operate a computer application. In this case we use EEG signals to control the position of a cursor. In this project healthy subjects learned to control the cursor in a trial-based setup. In offline analysis of the EEG data I tried to determine the target location for the cursor. The goal was to model and exploit time structure within a trial. I modelled the trial as a Markov process, using Hidden-Markov Models for classification. By using alternative training schemes for Markov models and comparing performance of these approaches, I will also make claims about the nature of the underlying time structure.