An Algorithm for Integer Least-squares with Equality, Sparsity and Rank Constraints

In this work, we deal with rank-constrained integer least-squares optimization problems arising in low-rank matrix factorization related applications. We propose a solution for constrained integer least-squares problem subject to equality, sparsity, and rank constraints. The algorithm combines the Fincke-Pohst enumeration (or sphere decoding algorithm) with rank constraints and sparse solutions of Diophantine equations to arrive at an optimal solution. The proposed approach consists of two steps as follows: (i) find the solution set for Diophantine equations arising from the linear and sparsity constraints, (ii) find the matrix which minimizes the integer least-squares objective and satisfying the rank constraints using the solution set obtained in the step 1. The proposed algorithm is illustrated using a simple example. Then, we perform experiments to study the computational aspects of different steps of the proposed algorithm.