# Robust Subspace-Constrained Quadratic Models for Low-Dimensional Structure Learning

Source: [Robust Subspace-Constrained Quadratic Models for Low-Dimensional Structure Learning](<https://www.sota2.com/research/paper/6a0e9190f984739ed6738aa6>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-05-22T05:00:05.162Z

Authors: Zheng Zhai, Xiaohui Li

Published: 2026-05-19T12:35:13.000Z

## Abstract

In this paper, we propose a robust subspace-constrained quadratic model (SCQM) for learning low-dimensional structure from high-dimensional data. Building upon the subspace-constrained quadratic matrix factorization (SQMF) framework, the proposed model accommodates a broad class of noise distributions, including generalized Gaussian and radial Laplace models. This generalization enables reliable performance under both heavy-tailed and light-tailed noise, thereby substantially enhancing robustness across diverse data regimes. To efficiently address the resulting nonconvex optimization problem, we develop a gradient-based algorithm equipped with a backtracking line-search strategy that ensures stable and efficient convergence. In addition, we present a sensitivity analysis of the $\\ell\_p^p$ and $\\ell\_2$ loss functions, elucidating their distinct behaviors under varying noise characteristics. Extensive numerical experiments corroborate the theoretical analysis and demonstrate that the proposed approach consistently outperforms existing methods in terms of robustness and reconstruction accuracy.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Manifold Denoising on 3D sphere dataset sigma=0.03](<https://www.sota2.com/research/sota/manifold-denoising-on-3d-sphere-dataset-sigma-0-03.md>): Manifold Denoising; 3D sphere dataset sigma=0.03; MSE: 0.001.

- [Manifold Denoising on 3D sphere dataset sigma=0.06](<https://www.sota2.com/research/sota/manifold-denoising-on-3d-sphere-dataset-sigma-0-06.md>): Manifold Denoising; 3D sphere dataset sigma=0.06; MSE: 0.005.

- [Manifold Denoising on 3D sphere dataset sigma=0.15](<https://www.sota2.com/research/sota/manifold-denoising-on-3d-sphere-dataset-sigma-0-15.md>): Manifold Denoising; 3D sphere dataset sigma=0.15; MSE: 0.053.

- [Manifold Denoising on 3D sphere dataset sigma=0.18](<https://www.sota2.com/research/sota/manifold-denoising-on-3d-sphere-dataset-sigma-0-18.md>): Manifold Denoising; 3D sphere dataset sigma=0.18; MSE: 0.082.

- [Manifold Denoising on 3D sphere dataset sigma=0.09](<https://www.sota2.com/research/sota/manifold-denoising-on-3d-sphere-dataset-sigma-0-09.md>): Manifold Denoising; 3D sphere dataset sigma=0.09; MSE: 0.015.

- [Manifold Denoising on 3D sphere dataset sigma=0.12](<https://www.sota2.com/research/sota/manifold-denoising-on-3d-sphere-dataset-sigma-0-12.md>): Manifold Denoising; 3D sphere dataset sigma=0.12; MSE: 0.02.

- [Manifold Denoising on 3D sphere dataset sigma=0.21](<https://www.sota2.com/research/sota/manifold-denoising-on-3d-sphere-dataset-sigma-0-21.md>): Manifold Denoising; 3D sphere dataset sigma=0.21; MSE: 0.125.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2605.20300>)

[PDF](<https://arxiv.org/pdf/2605.20300.pdf>)
