PASCAL - Pattern Analysis, Statistical Modelling and Computational Learning

Sparse NonGaussian Component Analysis, {\em IEEE Trans. on Inf. Theory
E. Diederichs, A. Juditsky and V. Spokoiny
IEEE Information Theory 2006.

Abstract

This article proposes a new approach to non-gaussian component analysis (NGCA). NGCA is already in use in high dimensional data analysis. It identifies non-gaussian components in the data as a preprocessing step for efficient information processing and is essentially based on Independent Component Analysis (ICA) and Principle Component Analysis (PCA). Instead we suggest an iterative and structure adaptive approach to non-gaussian component analysis (SNGCA) which is based on iterative convex projection. As an alternative to the use of principle components, SNGCA uses a statistical procedure which combines the computation of the Fritz John ellipsoid and statistical tests for normality as a tool for reducing the dimension of the data space.

EPrint Type:Article
Project Keyword:Project Keyword UNSPECIFIED
Subjects:Learning/Statistics & Optimisation
ID Code:5176
Deposited By:Anatoli Iouditski
Deposited On:24 March 2009