Title: Randomized Sub-Sampled Methods for Matrix Approximations
We introduce a family of randomized numerical algorithms which iteratively compute approximations of a matrix or its inverse or both. Our methods sample input matrices with respect to randomized input and output subspaces. Sub-sampling methods are similar to existing Quasi-Newton methods but operate with significantly less input data per iteration allowing better parallelism and scalability. Variants of our randomized sub-sampled methods are analyzed in various settings and compared to standard algorithms in literature.