Abstract

We consider the problem of learning a low-rank matrix, constrained to lie ina linear subspace, and introduce a novel factorization for modeling suchmatrices. A salient feature of the proposed factorization scheme is itdecouples the low-rank and the structural constraints onto separate factors. Weformulate the optimization problem on the Riemannian spectrahedron manifold,where the Riemannian framework allows to develop computationally efficientconjugate gradient and trust-region algorithms. Experiments on problems such asHankel matrix learning, non-negative matrix completion, and robust matrixcompletion demonstrate the efficacy of our approach.

 

Quick Read (beta)