SVD Factorization for Tall-and-Fat Matrices on Map/Reduce Architectures.
Burak Bayramli · arXiv (Cornell University) · 2013
We demonstrate an implementation for an approximate rank-k SVD factorization, combining well-known randomized projection techniques with previously implemented map/reduce solutions in order to compute steps of the random projection based SVD procedure, such QR and SVD. We structure the problem in a way that it reduces to Cholesky and SVD factorizations on $k \times k$ matrices computed on a single machine, greatly easing the computability of the problem.