一种分离低维信号的ICA快速算法
doi: 10.3724/SP.J.1004.2011.00794
A Simple and Accurate ICA Algorithm for Separating Mixtures of Up to Four Independent Components
-
摘要: 介绍了一种基于低维反对称矩阵指数的快速独立分量分析算法. 由于算法中牵涉到的矩阵指数具有解析闭合形式的表达, 因而算法中使用到的矩阵指数以及最优下降方向均可解析地得到. 另外, 我们纠正了在别的文献中建立的四维反对称矩阵指数表达式中的两个错误. 最后, 我们用仿真验证了算法. 实验结果表明: 相比于广为应用的Extended InfoMax和FastICA算法, 本文算法能得到更佳的分离性能.Abstract: This paper introduces an algorithm for independent component analysis (ICA) using explicit closed forms of two-, three- and four-dimensional antisymmetric matrix exponentials, based on which both the search direction and matrix exponentials can be directly computed in each iteration without any approximation. In addition, two errors have been corrected for the representation of four-dimensional antisymmetric matrix exponentials that were established in other works. Simulations show that the algorithm converges fast and can achieve better performance than the well-known Extended InfoMax and FastICA algorithms for mixtures of up to four independent components.
-
Key words:
- Independent component analysis (ICA) /
- matrix exponential /
- closed form
-
[1] Choi S, Cichocki S, Park H M, Lee S Y. Blind source separation and independent component analysis: a review. Neural Information Processing-Letters and Reviews, 2005, 6(1): 1-57 [2] Xiao Ming, Xie Sheng-Li, Fu Yu-Li. Underdetermined blind source separation algorithm based on normal vector of hyperplane. Acta Automatica Sinica, 2008, 34(2): 142-149(in Chinese)[3] Tang Ying, Li Jian-Ping. A new algorithm of ICA: using the parameterized orthogonal matrixes of any dimensions. Acta Automatica Sinica, 2008, 34(1): 31-39(in Chinese)[4] Bell A J, Sejnowski T J. An information-maximization approach to blind separation and blind deconvolution. Neural Computation, 1995, 7(6): 1129-1159[5] Ashi H A, Cummings L J, Matthews P C. Comparison of methods for evaluating functions of a matrix exponential. Applied Numerical Mathematics, 2009, 59(3-4): 468-486[6] Abrudan T E, Eriksson J, Koivunen V. Steepest descent algorithms for optimization under unitary matrix constraint. IEEE Transactions on Signal Processing, 2008, 56(3): 1134-1147[7] Fiori S. Quasi-geodesic neural learning algorithms over the orthogonal group: a tutorial. Journal of Machine Learning Research, 2005, 6: 743-781[8] Fiori S, Tanaka T. An algorithm to compute averages on matrix lie groups. IEEE Transactions on Signal Processing, 2009, 57(12): 4734-4743[9] Nishimori Y, Akaho S. Learning algorithms utilizing quasi-geodesic flows on the stiefel manifold. Neurocomputing, 2005, 67: 106-135 [10] Plumbley M D. Algorithms for nonnegative independent component analysis. IEEE Transactions on Neural Networks, 2003, 14(3): 534-543[11] Gallier J, Xu D. Computing exponentials of skew-symmetric matrices and logarithms of orthogonal matrices. International Journal of Robotics and Automation, 2002, 17(4): 10-20[12] Moakher M. Means and averaging in the group of rotations. SIAM Journal on Matrix Analysis and Applications, 2002, 24(1): 1-16 [13] Politi T. A formula for the exponential of a real skew-symmetric matrix of order 4. Bit Numerical Mathematics, 2001, 41(4): 842-845[14] Pham D T, Garrat P. Blind separation of mixture of independent sources through a quasi-maximum likelihood approach. IEEE Transactions on Signal Processing, 1997, 45(7): 1712-1725[15] Xu L. One-bit-matching theorem for ICA, convex-concave programming on polyhedral set, and distribution approximation for combinatorics. Neural Computation, 2007, 19(2): 546-569[16] Lee T W, Girolami M, Sejnowski T J. Independent component analysis using an extended infomax algorithm for mixed subgaussian and supergaussian sources. Neural Computation, 1999, 11(2): 417-441[17] Shen H, Kleinsteuber M, Huper K. Local convergence analysis of fastICA and related algorithms. IEEE Transactions on Neural Networks, 2008, 19(6): 1022-1032[18] Gloub G H, Loan C F V. Matrix Computation (Third Edition). Baltimore: The John Hopkins University Press, 1996. 341-342
点击查看大图
计量
- 文章访问数: 1899
- HTML全文浏览量: 54
- PDF下载量: 889
- 被引次数: 0