project summary
This project provides a matlab implementation of several fixed point algorithms for performing independent component analysis both in a noise free setting and in the presence of an additive Gaussian noise. In particular, this includes the gradient iteration ICA (GI-ICA) algorithm described in the paper "Fast Algorithms for Gaussian Noise Invariant Independent Component Analysis" by James Voss, Luis Rademacher, and Mikhail Belkin as well as the pseudo-Euclidean gradient iteration (PEGI) algorithm described in the paper "A Pseudo-Euclidean Iteration for Optimal Recovery in Noisy ICA" by James Voss, Luis Rademacher, and Mikhail Belkin.
downloads
The current version of the GI-ICA code is available for download here:
The current version is also available on our sourceforge website http://sourceforge.net/projects/giica/.
code usage conditions
This GI-ICA project is a matlab implementation of the GI-ICA algorithm for Independent Component Analysis designed to be robust to Gaussian noise as described in the paper:
This code is made available under the GPLv3 license. We ask that any academic work which uses this code cite the most appropriate of the aforementioned papers.
using GI-ICA
The code should be kept together in a single directory in order to satisfy all interdependencies. The file demo_giica.m provides a basic demo showing how to run the GI-ICA algorithm on simulated data. For running the GI-ICA algorithm on your own data, the main function to interface with is in GIICA.m. The main function for the gradient-iteration ICA algorithms is actually implemented in the file ICA_Implementation.m.
We refer the user to the comments at the top of the file GIICA.m for a detailed explanation of all usage options when running the GI-ICA algorithm.