Workshop on Structure Adapting Methods - Abstract

Diederichs, Elmar

Sparse NonGaussian Component Analysis by Semidefinite Programming

Sparse NonGaussian component analysis is an unsupervised, linear, iterative and structure adaptive extraction method for semiparametric high dimensional data analysis based on estimating low-dimensional non-Gaussian components of the high-dimensional data. We propose a simplified approach to that method based on semidfinite programming that allows to improve the statistical sensitivity to a broad variety of devitations from normality while decreasing its computational effort. The new approach is demonstrated an high dimensional data sets obtained from MD-simulations of proteins.