ABSTRACT
As a noninvasive neuroimaging technique, electroencephalogram (EEG) has been most popularly applied to brain–computer interface (BCI) development. Informative feature extraction and accurate classification of EEG patterns are considerably challenging because of the poor signal-to-noise ratio caused by volume conduction effects, interferences from various noises, and intrinsic signal nonstationarity. In this chapter, we provide a tutorial of applying Bayesian learning-based algorithms to EEG feature optimization and classification in BCI applications. The algorithm principles of Bayesian learning are detailedly explained in the aspects of prior designing, posterior inference, and hyperparameter estimation. With two representative examples based on event-related potential and sensorimotor rhythm BCI paradigms, we further introduce the usage of Bayesian learning-based algorithms for EEG analysis. Extensive experimental comparisons are carried out among different competing algorithms to validate the effectiveness of Bayesian learning in BCI applications. A discussion is provided on various extensions of Bayesian learning for exploring more complex but potentially important properties of EEG, toward improved performance of BCI systems.
