Dimensionality Reduction in ICA and Rank-(R_1, R_2, ... , R_N) Reduction in Multilinear Algebra

摘要

We show that the best rank-(R_1, R_2, ... , R_n) approximation in multilinear algebra is a powerful tool for dimensionality reduction in ICA schemes without prewhitening. We consider the application to different classes of ICA algorithms.