Mining Structured Sets Of Subspaces From High Dimensional Data