Tucker Tensor Decomposition of Multi-session EEG Data

Abstract

The Tucker model is a tensor decomposition method for multi-way data analysis. However, its application in the area of multi-channel electroencephalogram (EEG) is rare and often without detailed electrophysiological interpretation of the obtained results. In this work, we apply the Tucker model to a set of multi-channel EEG data recorded over several separate sessions of motor imagery training. We consider a three-way and four-way version of the model and investigate its effect when applied to multi-session data. We discuss the advantages and disadvantages of both Tucker model approaches.