中文

Implicit Regularization in Perturbed Deep Matrix Factorization: Spectral Conditions and Stability

最优化与控制 2026-05-28 v1 机器学习 机器学习

摘要

This paper studies the stability of low-rank implicit regularization in perturbed deep matrix factorization, where the target matrix is corrupted by a noise matrix. We first derive sufficient spectral conditions under which gradient descent exhibits a low-rank phase in the noiseless setting. These conditions show how the target spectrum, initialization, and step size jointly determine the existence of a nonempty low-rank interval. We then analyze the perturbed gradient descent dynamics, proving convergence guarantees and quantifying how the perturbation affects iteration complexity and eigenvalue recovery. Finally, we show that the low-rank phase persists under perturbation, with explicit dependence on the perturbation size. Numerical experiments support the theoretical findings.

关键词

引用

@article{arxiv.2605.28613,
  title  = {Implicit Regularization in Perturbed Deep Matrix Factorization: Spectral Conditions and Stability},
  author = {Jingzhe Wang and Hung-Hsu Chou},
  journal= {arXiv preprint arXiv:2605.28613},
  year   = {2026}
}