Exploiting Ongoing EEG With Multilinear Partial Least Squares During Free-Listening To Music
Deqing Wang, Dalian University of Technology
Fengyu Cong, Dalian University of Technology
Qibin Zhao, RIKEN Brain Science Institute
Petri Toiviainen, University of Jyväskylä
Asoke K Nandi, Brunel University
Minna Huotilainen, Finnish Institute of Occupational Health
Tapani Ristaniemi, University of Jyväskylä
Andrzej Cichocki, RIKEN Brain Science Institute

Abstract:
During real-world experiences, determining the stimulus-relevant brain activity is excitingly attractive and is very challenging, particularly in electroencephalography. Here, spectrograms of ongoing electroencephalogram (EEG) of one participant constructed a third-order tensor with three factors of time, frequency and space; and the stimulus data consisting of acoustical features derived from the naturalistic and continuous music formulated a matrix with two factors of time and the number of features. Thus, the multilinear partial least squares (PLS) conforming to the canonical polyadic (CP) model was performed on the tensor and the matrix for decomposing the ongoing EEG. Consequently, we found that brain activity of majority of participants was significantly correlated with the musical features in time domain, and that such brain activity showed frontal or central or posterior or occipital distributions along the scalp, and that such brain activity could be of different oscillation bands in frequency domain.