149ad6e32c08b73a3ecc3d11977fcc47-Paper-Conference.pdf

We propose a regularized pairwise pseudo-likelihood approach for matrix completion and provethat the proposed estimator can asymptotically recoverthe low-rank parameter matrix uptoanidentifiable equivalence class of aconstant shiftandscaling, atanear-optimal asymptotic convergencerateofthe standardwell-posed(non-informativemissing)setting,whileeffectivelymitigating the impact of informative missingness.