شماره ركورد كنفرانس :
3140
عنوان مقاله :
ICA for heavy - tail a — stable sources
عنوان به زبان ديگر :
ICA for heavy - tail a — stable sources
پديدآورندگان :
Shokripour Mona نويسنده Department of Statistics - Shahid Beheshti University - Tehran - Iran , Mohammadpouro Adel نويسنده Department of Statistics - Amirkabir University of Technology - Tehran - Iran
كليدواژه :
Discrete spectral measure , ICA , Independent Component Analysis , Cy-stable class of distributions , Sub-Gaussian class of distributions
عنوان كنفرانس :
يازدهمين كنفرانس آمار ايران
چكيده لاتين :
Abstract. Most of the classical solutions of independent component analysis are failed to use when we face with heavy tail observations with infinite second moment. Cy-stable class of distributions is a famous class of heavytail distributions which dealing with them is difficult because of their special properties. An efficient method proposed in this paper to solve the ICA problem for the case of C-stable sources, is based on the estimation of discrete spectral measure of an (n-stable random vector. We also study a special case of Sub-Gaussian class, where ICA does not exist. Some discussions and simulations are also provided.
شماره مدرك كنفرانس :
4219389
