On vector averaging over the unit hypersphere

Sample averaging is a commonly used way to smooth out irregularities of data and to get rid of random fluctuations in measurements analysis. In adaptive signal processing, where an adaptive system learns its own parameters in order to perform a predefined task, the learnt parameters-pattern may depend on the initial learning state and on the fluctuations of the statistical features of the input signals to the system. In adaptive system learning, averaging may be employed as a method to merge several learnt parameters-patterns in order to get a better representative pattern. Even in the case of scalar parameters, the concept of averaging is not uniquely defined as scalar parameters spaces may exhibit a rich structure to be dealt with. The case of multiple parameter patterns where single parameters are mutually constrained to each other may exhibit an even richer structure. In the present paper, we deal with the case of parameters-patterns belonging to the unit hypersphere and develop an averaging technique based on the differential geometrical structure of such a curved space. Numerical experiments illustrate the behavior of the developed averaging algorithm.

[1]  Simone G. O. Fiori,et al.  One-unit 'rigid-bodies' learning rule for principal/independent component analysis with application to ECT-NDE signal processing , 2004, Neurocomputing.

[2]  Inderjit S. Dhillon,et al.  Clustering with Bregman Divergences , 2005, J. Mach. Learn. Res..

[3]  Aapo Hyvärinen,et al.  Fast and robust fixed-point algorithms for independent component analysis , 1999, IEEE Trans. Neural Networks.

[4]  Zhi Ding,et al.  Blind Equalization and Identification , 2001 .

[5]  R. Sundaram.  A First Course in Optimization Theory: Bibliography , 1996 .

[6]  Ya-lun Chou,et al.  Statistical analysis, with business and economic applications , 1975 .

[7]  M. Fréchet.  Les éléments aléatoires de nature quelconque dans un espace distancié , 1948 .

[8]  L. Lopez,et al.  Runge-Kutta Type Methods Based on Geodesics for Systems of ODEs on the Stiefel Manifold , 2001 .

[9]  R. Wiggins.  Minimum entropy deconvolution , 1978 .

[10]  Simone G. O. Fiori,et al.  Geodesic-based and projection-based neural blind deconvolution algorithms , 2008, Signal Process..

[11]  Roger Koenker.  THE MEDIAN IS THE MESSAGE : TOWARD THE FRÉCHET MEDIAN , 2007 .

[12]  Ye Li,et al.  Blind Channel Equalization and Identification , 2001 .

[13]  Simone Fiori,et al.  Neural minor component analysis approach to robust constrained beamforming , 2003 .

[14]  Ehud Weinstein,et al.  Super-exponential methods for blind deconvolution , 1993, IEEE Trans. Inf. Theory.

[15]  Simone G. O. Fiori,et al.  Fast fixed-point neural blind-deconvolution algorithm , 2004, IEEE Transactions on Neural Networks.

[16]  Simone G. O. Fiori,et al.  Overview of independent component analysis technique with an application to synthetic aperture radar (SAR) imagery processing , 2003, Neural Networks.

[17]  Daesik Hong,et al.  Performance Of Neural Equalizers On Partial Erasure Model , 1997 .

[18]  L. Bregman.  The relaxation method of finding the common point of convex sets and its application to the solution of problems in convex programming , 1967 .