B0793
Title: Iterative regularization methods for ill-posed generalized linear models
Authors: Gianluca Finocchio - University of Vienna (Austria) [presenting]
Tatyana Krivobokova - University of Vienna (Austria)
Abstract: The problem of regularized maximum-likelihood estimation in ill-posed generalized linear models is studied. Ill-posedness is assumed to be the byproduct of a low-dimensional latent factor model. We provide a class of iterative algorithms extending known penalization/projection techniques and obtain theoretical guarantees under regularity assumptions on the latent model. In particular, when the number of features and observations are both large, we propose a novel iteratively-reweighted-partial-least-squares algorithm outperforming its competitors in both computational efficiency and minimum-norm maximum-likelihood estimation. Our findings are confirmed by simulation studies on both real and generated data.
