Analysis of Brain Data - Using Multi-Way Array Models on the EEG

Morten Mørup

AbstractIn this thesis the multi-way array model Parallel Factors (PARAFAC) also known as Canonical Decomposition (CANDECOMP) was applied to the event related potential (ERP) of electroencephalographic (EEG) recordings. Previous work done analyzing the ERP by PARAFAC had encountered great problems of degeneracy in the factors. However, in this thesis it is shown that the problem of degeneracy can be effectively circumvented by imposing non-negativity. Furthermore, the PARAFAC analysis was, to my knowledge, for the first time used to analyze the wavelet transformed data of the ERP. Through this analysis, it was shown that PARAFAC is able to access the correct components of the data. Finally, a novel PARAFAC algorithm based on independent component analysis on data having the concept of Combined Independence was proposed. This algorithm proved both fast and efficient in accessing the correct components of simulated as well as real data. In dealing with noise, the algorithm performed even better than the popular PARAFAC algorithm based on alternating least squares.
KeywordsPARAFAC, CANDECOMP, ERP, EEG, coherence, ITPC, multi-way arrays, tensors, Independent Component Analysis, gamma activity, wavelet analysis, Combined Independence, HOSVD, TUCKER, Core Consistency Diagnostic
TypeMaster's thesis [Academic thesis]
PublisherInformatics and Mathematical Modelling, Technical University of Denmark, DTU
AddressRichard Petersens Plads, Building 321, DK-2800 Kgs. Lyngby
NoteSupervised by Prof. Lars Kai Hansen
Electronic version(s)[pdf]
BibTeX data [bibtex]
IMM Group(s)Intelligent Signal Processing

Back  ::  IMM Publications