The EEG is a recording usually comprising multiple channels each containing
mixed versions of underlying brain sources along with artifacts. Most brain
signal analysis systems rely on extracting a signal of interest, whether in
clinical data (e.g. epileptic spikes and seizures) or in BCI. Most analysis
techniques usually comprise an artifact rejection scheme followed by a
feature extraction then by classification (in a classification problem).
Blind Source Separation offers a unique de-noising scheme - it separates out
the mixed recordings into their underlying components or sources in a
physiologically meaningful way. This means that such schemes can remove
artifacts and isolate sources of interest using the same analysis criteria.
Most commonly, ICA is used as the BSS technique of choice.

This talk will introduce ICA and extensions to ICA that are useful for EEG
denoising and subsequent analysis. Clinical EEG examples will be shown, and
others - which specifically address the issues faced in practical BCI - i.e.
BCI that can be used outside the laboratory.