On robust spatial filtering of EEG in nonstationary environments

Brain-Computer Interfacing (BCI) is a promising technology for patients that are severely motor-disabled, because it enables them to communicate and interact with the environment. A BCI system decodes user's intentions from brain signals, typically recorded with electroencephalography (EEG), and transmits them to a computer application that, e.g., controls a wheelchair. The efficiency of the system largely depends upon a reliable extraction of informative features from the high-dimensional EEG signal. Spatial filtering is a crucial step in this protocol, however, current approaches are prone to errors when data is contaminated by artifacts or is nonstationary. This article provides an overview of a dissertation, which has addressed the problem of robust spatial filtering in BCI. The contributions of the thesis range from the development of regularization schemes and a robust parameter estimator for spatial filtering, to the formulation of an information geometric view on the spatial filtering problem and the proposal of a new family of algorithms based on robust divergences. The developed methods and concepts are applicable to a variety of problems in machine learning and signal processing.