Extracting a Desired Independent Source in ICA

摘要

We propose a novel approach to extract a desired independent source from multidimensional observations when priori information of the source is available. An extension to the constrained independent component analysis (cICA) is adopted to model such extraction. A solution to the constrained optimization problem with a Newton-like learning algorithm is provided. The convergence of the learning procedure is proved and analyzed. Simulations of extracting a single source from Gaussian-class signals and an activation time response from synthetic fMRI data demonstrate the efficacy and accuracy of our algorithm compared to other methods.