The computation of the leading part of the singular value decomposition (SVD) of a matrix is an important problem in numerical linear algebra. In applications SVDs appear in dimension-reduction reduction techniques (principal component analysis (PCA), etc.), and the matrices are often very large, and in some applications also dense and distributed over a network of machines. We are interested in computational approaches that deal specifically with the distributed nature of the available computational resources, which may run at very different speeds and may be prone to failure, making it necessary to develop methods that don't rely on synchronous communication between the computational nodes.
