The singular value decomposition (SVD) technique is extended to incorporate the additive components for approximation of a rectangular matrix by the outer products of vectors. While dual vectors of the regular SVD can be expressed one via linear transformation of the other, the modified SVD corresponds to the general linear transformation with the additive part. The method obtained can be related to the family of principal component and correspondence analyses, and can be reduced to an eigenproblem of a specific transformation of a data matrix. This technique is applied to constructing dual eigenvectors for data visualizing in a two dimensional space.
"Generalized Singular Value Decomposition with Additive Components,"
Journal of Modern Applied Statistical Methods:
1, Article 29.
Available at: http://digitalcommons.wayne.edu/jmasm/vol15/iss1/29