29
0

An extrapolated and provably convergent algorithm for nonlinear matrix decomposition with the ReLU function

Abstract

Nonlinear matrix decomposition (NMD) with the ReLU function, denoted ReLU-NMD, is the following problem: given a sparse, nonnegative matrix XX and a factorization rank rr, identify a rank-rr matrix Θ\Theta such that Xmax(0,Θ)X\approx \max(0,\Theta). This decomposition finds application in data compression, matrix completion with entries missing not at random, and manifold learning. The standard ReLU-NMD model minimizes the least squares error, that is, Xmax(0,Θ)F2\|X - \max(0,\Theta)\|_F^2. The corresponding optimization problem is nondifferentiable and highly nonconvex. This motivated Saul to propose an alternative model, Latent-ReLU-NMD, where a latent variable ZZ is introduced and satisfies max(0,Z)=X\max(0,Z)=X while minimizing ZΘF2\|Z - \Theta\|_F^2 (``A nonlinear matrix decomposition for mining the zeros of sparse data'', SIAM J. Math. Data Sci., 2022). Our first contribution is to show that the two formulations may yield different low-rank solutions Θ\Theta; in particular, we show that Latent-ReLU-NMD can be ill-posed when ReLU-NMD is not, meaning that there are instances in which the infimum of Latent-ReLU-NMD is not attained while that of ReLU-NMD is. We also consider another alternative model, called 3B-ReLU-NMD, which parameterizes Θ=WH\Theta=WH, where WW has rr columns and HH has rr rows, allowing one to get rid of the rank constraint in Latent-ReLU-NMD. Our second contribution is to prove the convergence of a block coordinate descent (BCD) applied to 3B-ReLU-NMD and referred to as BCD-NMD. Our third contribution is a novel extrapolated variant of BCD-NMD, dubbed eBCD-NMD, which we prove is also convergent under mild assumptions. We illustrate the significant acceleration effect of eBCD-NMD compared to BCD-NMD, and also show that eBCD-NMD performs well against the state of the art on synthetic and real-world data sets.

View on arXiv
@article{gillis2025_2503.23832,
  title={ An extrapolated and provably convergent algorithm for nonlinear matrix decomposition with the ReLU function },
  author={ Nicolas Gillis and Margherita Porcelli and Giovanni Seraghiti },
  journal={arXiv preprint arXiv:2503.23832},
  year={ 2025 }
}
Comments on this paper