ABSTRACT
The singular value decomposition is best thought of as a transformation in a geometric space. The SemiDiscrete Decomposition (SDD), although it was originally developed as a space-efficient version of SVD, is best thought of as working with the entries of the matrix directly. It searches for regions in the matrix that have entries of roughly similar magnitude, and treats each of these as a component. In many situations, this is a useful way to decompose the matrix.
