An Exterior Method for Nonnegative Matrix Factorization

TL;DR

This paper introduces an exterior framework for nonnegative matrix factorization (NMF) that improves convergence and solution quality by separating low-rank approximation from nonnegativity constraints, outperforming existing methods.

Contribution

The proposed eNMF method initializes from unconstrained solutions and employs a rotation procedure to efficiently find boundary solutions, offering a new geometric perspective and practical advantages.

Findings

eNMF outperforms 81 state-of-the-art algorithms in reconstruction error and speed.

In 99% of experiments, different algorithms converge to equivalent solutions.

Downstream tasks show significant performance improvements with eNMF.

Abstract

Nonnegative matrix factorization (NMF) seeks a low-rank approximation $X \approx UV^T$ with nonnegative factors and is commonly solved using interior methods that enforce feasibility throughout optimization. We show that such constraint-driven approaches can impede progress in the nonconvex landscape, leading to slow convergence or convergence to suboptimal stationary points. We propose an exterior framework for NMF (eNMF) that separates low-rank approximation from nonnegativity enforcement. Our method initializes from the optimal unconstrained factorization and introduces a rotation procedure that maps unconstrained factors to an exterior point closest to the nonnegative orthant. This viewpoint yields an algorithmic framework in which simple iterative updates converge to KKT-satisfying stationary points on the boundary of the positive orthant. The exterior formulation also enables a geometric interpretation of NMF solutions, clarifying equivalence classes of factorizations under permutation and orthogonal transformations. An intriguing numerical result, involving 400 NMF experiments across both real and synthetic datasets, show that in 99% of the cases, different algorithms tend to converge towards equivalent factor matrices. We benchmark eNMF against 9 state-of-the-art NMF algorithms with 9 initialization schemes across 3 real-world and 2 synthetic datasets. eNMF consistently outperforms all 81 competitors, achieving up to 30% lower reconstruction error under equal-time settings and up to 150% speedup under equal-error settings. The downstream experiments further demonstrate substantial performance gains in audio processing and recommendation tasks, corroborating the practical benefits of the proposed exterior optimization framework. Code is available at https://github.com/roychowdhuryresearch/eNMF