A Convex Cauchy-Schwarz Divergence Measure for Blind Source Separation


We propose a new class of divergence measures for Independent Component Analysis (ICA) for the de-mixing of sources.  The Convex Cauchy-Schwarz Divergence (CCS-DIV) is formed by integrating convex functions into the Cauchy-Schwarz inequality.  The new measure is symmetric and convex, where the degree of convexity can be tuned by a (convexity) parameter.  The CCS-DIV is more likely to attain optimal minima and speedup the search process in ICA. A non-parametric algorithm, generated from the proposed divergence, is developed with convexity parameters and is employing the Parzen window-based distribution.

  1. Zaid Albataineh and Fathi Salem, “Convex Cauchy-Schwarz Independent Component Analysis CCS-ICA for Blind Source Separation,” IEEE Trans. Neural Netw. 2014
  2. Zaid Albataineh and Fathi Salem, “RobustICA-Based Algorithm for Blind Separation of Convolutive Mixtures” IEEE Trans. Circuit and System I 2014
  3. Zaid Albataineh and Fathi Salem, “A Convex Cauchy-Schwarz Divergence Measure for Blind Source Separation” Springer. Neural Processing Letter 2014
  4. Zaid Albataineh and Fathi Salem, “Two Pairwise Iterative Schemes For High Dimensional Blind Source Separation,” Springer. Neural Processing Letter 2014