CFE 2019: Start Registration
View Submission - CMStatistics
Title: Asymptotically and computationally efficient tensorial JADE Authors:  Joni Virta - Aalto University (Finland) [presenting]
Niko Lietzen - Aalto University School of Science (Finland)
Pauliina Ilmonen - Aalto University School of Science (Finland)
Klaus Nordhausen - Vienna University of Technology (Austria)
Abstract: A novel method of tensorial independent component analysis is proposed based on TJADE and $k$-JADE, two recently proposed generalizations of the classical JADE algorithm. The new method achieves the consistency and the limiting distribution of TJADE under mild assumptions, and at the same time, it offers notable improvement in computational speed. The trade-off between computational speed and assumptions is controlled by a tuning parameter which has a natural interpretation as a maximal kurtosis multiplicity. Simulations and timing comparisons demonstrate the method's gain in speed and, moreover, the desired efficiency is obtained approximately also for finite samples. The method is applied successfully to large-scale video data, for which neither TJADE nor $k$-JADE is feasible.