Abstract
In this talk we will review several promising paradigms for Brain Computer Interface, (including P300/N170 ERPs, SSVEP, and motor imagery-MI paradigms) and multi-way (tensor) signal processing tools for EEG-BCI and analysis of brain to brain couplings/interactions (BBC/I). We will discuss how tensor (multiway arrays) factorizations/decompositions can be applied for classification and recognition of evoked and event related potentials (EP/ERP). We illustrate this by Multiway Canonical Correlation Analysis (MCCA) which is applied to improve recognition rate of the Steady State Visual Evoked Potentials (SSVEP). Furthermore, we will present affective brain-computer interfaces (aBCI) based on oddball paradigm using visual stimuli with emotional facial images and short video-clips. Our experiments confirmed that the face-sensitive event-related potential (ERP) components N170 and vertex positive potentials (VPP) have reflected early structural encoding of emotional faces and allows us to improve performance and reliability of BCI. The developed multiway (tensor) signal processing tools are promising not only for BCI but also for real time neurofeedback (NF) and EEG hyper-scanning to investigate human emotions, social interactions and brain to brain couplings/interactions. Dynamic tensor analysis allows us to discover meaningful hidden structures of complex brain data and to extract hidden components or features by capturing multi-linear and multi-aspect relationships. The challenge is how to analyze intractably large-scale brain data for such problems as dimensionality reduction, feature extraction, classification, clustering and anomaly detection.
