Signal processing for EEG-based brain-computer interfaces

Electroencephalography (EEG)-based brain-computer interface (BCI) offers advantages in terms of temporal resolution, cost, and mobility, and is therefore the most popular non-invasive BCI modality to date. However, the current paradigm has a narrow application scope as it relies heavily on synthetic stimuli, multi-trial averaging techniques and the active participation of subjects. This project takes a step forward to real-world settings, aiming to identify and quantify the temporal coupling between natural video clips and elicited EEG responses. We will explore multi-set canonical correlation analysis and its nonlinear extensions to enhance the EEG signals and link them with (generic) stimulus features. In collaboration with researchers in the PSI group, we aim to find a good deep joint video-EEG embedding and perform visual attention decoding.