PCA-ICA-PFA - combinatie van principale componenten analyse, onafhankelijke componentenanalyse en parallelle factor analyse

From 01-10-2008 to 31-03-2013

Description

The goal of this project is to combine the analysis techniques mentioned above. Let us illustratethe principle by means of an example. In DS-CDMA telecommunication systems, the signals ofthe different users can be separated by means of CPA [44]. Here one does not impose orthogonalityconstraints on the CP decomposition. Nevertheless it is straightforward to assume that thesignals of the different users are uncorrelated. This hypothesis can be incorporated by imposingorthogonality in one “mode” of the CP decomposition. Although, strictly speaking, this extraconstraint is not necessary for the uniqueness of the result, it will probably lead to a higher accuracyand a greater stability of the solution. One can go one step further and claim that the signalsof the different users are statistically independent, like in ICA. One then imposes statistical independence,instead of orthogonality, in one mode of the CP decomposition. The same principleapplies in many cases to the separation of biomedical signals (e.g. EEG, epilepsy). Recently,CPA has been used as a way to drastically increase the speed of high-dimensional computations[9, 10, 27, 32]. Here, orthogonality in one or several modes could be desirable from a numericalpoint of view.

Team

• Lieven De Lathauwer, Promoter
• Sabine Van Huffel, Team member

Financing

Funding: KU Leuven - Internal Funding KU Leuven

Program/Grant Type: BOF STRT - KUL Starting Grant