- Abstract: Electroencephalography-based brain-computer interfaces are systems that infer brain signals recorded using electroencephalography (EEG) to provide a means of communication for patients suffering from locked-in syndrome where common neuromuscular pathways are not available. One challenge in EEG-based BCIs is non-stationarity of the EEG signal. A major contributor to this is feedback-related brain activity. Since EEG consists of time series data recorded at multiple sites on the scalp, one can estimate covariance matrices for both time and space which lie on the Riemannian manifold of symmetric positive definite matrices. In this work, we investigate spatio-temporal aspects of the feedback-related brain activity by considering both space and time covariances in Euclidean space and on the Riemannian manifold. We propose two novel methods to incorporate both spatial and temporal features and show improved results compared to existing methods.
- Keywords: Brain-computer interfaces, Electroencephalography, Feedback-related brain activity, Riemannian manifold, Spatial and temporal features