Eigen and subspace updating with forward-backward averaging
R.D. DeGroat, Darel A. Linebarger · 2003
The authors exploit the structure of forward-backward (FB) matrices to achieve reduced computation. In particular, computationally efficient eigenvalue decomposition (EVD) and EVD updating methods are developed for FB correlation matrices. Similarly, efficient singular value decomposition (SVD) and SVD updating methods are developed for FB data matrices. A general approach to reducing the computation associated with an FB matrix is described.>