Autors: Constantin,D.
Title: Principal Directions for Local Independent Components Analysis
Keywords: Nonlinear Independent Component Analysis, Principal Componen

Abstract: In nonlinear independent component analysis, usually a series of difficulties related the parameter estimation arise. The reported work proposes an estimation method based on the use of the principal components and on the use of the algorithms of the linear ICA pattern. The estimation of the nonlinear ICA involves a local analysis pattern using the principal directions for the classification of input data and the application of the linear ICA pattern to the level of each class of elements. We use a classification part that corresponds to the nonlinear representation of the mixed data as well as a part of local application of the linear ICA pattern in order to describe the independent characteristics of the data. The purpose is to obtain better data representation as compared to the methods corresponding to the linear ICA pattern at global level. The proposed algorithm was tested in several signals separation applications and the obtained results prove good recognition performances.

References

    Issue

    9th WSEAS International Conference on NEURAL NETWORKS (NN’08), pp. 127-130, 2008, Bulgaria,

    Вид: публикация в национален форум