General Description
This software implementation included here is related to our paper presented at a conference [GaKi06]. This paper presents a Matlab implementation of the β-divergence for blind source separation algorithm by Mihoko and Eguchi [MiEg02].
The implementation uses a quasi-Newton Broyden-Fletcher-Goldfarb-Shanno (BFGS) optimization, combined with an Armijo conditioned line-search, to minimize the β-divergence between the density of the source estimates and the product of its hypothesized marginal densities to separate a mixture of statistically independent sources.
In addition, the implementation repeats simulations conducted by Mihoko and Eguchi, and uses the Amari separation performance index to provide a quantitative separation performance.
This zip file contains the Matlab code used to generate the results for the paper.
References
- [Fish04]
- J. Fisher. ICA data (Online as of February 13, 2004)
http://www.ai.mit.edu/people/fisher/ica_data/
- [GaKi06]
- N. Gadhok and W. Kinsner, "An implementation of beta-divergence for blind source separation", in Proc. IEEE 2006 Canadian Conference on Electrical and Computer Engineering, CCECE06 (Ottawa, Ontario; May 7-10, 2006) pp. 643-646 (4 pp.), 2006.
- [MiEg02]
- M. Mihoko and S. Eguchi, "Robust blind source separation by beta divergence," Neural Computation, vol. 14, pp. 1859-1886, 2002.
Code Description
File Name
- CCECE06_ICA_Beta_v10.zip
(Contains 25 .m files and 2 subdirectories)
Version
- 1.0
Subdirectories
- Fisher_Data (6 .m files, 4 subdirectories, 3 html files)
- genHyper_v1.1 (1 .m file, 1 .f90 file, 1 .mexglx file, and 1 readme)
Notes
- Internal Versions: Each .m file has a version listed with the file.
- BetaD_Caller_DataSet1_Iterative.m,
BetaD_Caller_DataSet2_Iterative.m, and
BetaD_Caller_DataSet3_Iterative.m are the primary caller programs for generating the results contained in [GaKi06].
- A number of included .m files are used to generate plots similar to [MiEg02], but were not published in [GaKi06].
- Synthetic ICA data generated by J. Fisher is included in the zip file [Fish04]. This is the same data used in [MiEg02].
Feedback
- Please provide any feedback.
Legal Notes
- This software is provided for research purposes only.
- No commercial use in any form is permitted.
- WE MAKE NO WARRANTIES, EITHER EXPRESS OR IMPLIED, REGARDING THE ENCLOSED COMPUTER SOFTWARE PACKAGE.
Code Downloading
Unzip the file after downloading.
- CCECE06_ICA_Beta_v10.zip (zip; 10.890 MB)
|