English

Model-free filtering in high dimensions via projection and score-based diffusions

Statistics Theory 2025-10-28 v1 Machine Learning Statistics Theory

Abstract

We consider the problem of recovering a latent signal XX from its noisy observation YY. The unknown law PX\mathbb{P}^X of XX, and in particular its support M\mathscr{M}, are accessible only through a large sample of i.i.d.\ observations. We further assume M\mathscr{M} to be a low-dimensional submanifold of a high-dimensional Euclidean space Rd\mathbb{R}^d. As a filter or denoiser X^\widehat X, we suggest an estimator of the metric projection πM(Y)\pi_{\mathscr{M}}(Y) of YY onto the manifold M\mathscr{M}. To compute this estimator, we study an auxiliary semiparametric model in which YY is obtained by adding isotropic Laplace noise to XX. Using score matching within a corresponding diffusion model, we obtain an estimator of the Bayesian posterior PXY\mathbb{P}^{X \mid Y} in this setup. Our main theoretical results show that, in the limit of high dimension dd, this posterior PXY\mathbb{P}^{X\mid Y} is concentrated near the desired metric projection πM(Y)\pi_{\mathscr{M}}(Y).

Keywords

Cite

@article{arxiv.2510.23197,
  title  = {Model-free filtering in high dimensions via projection and score-based diffusions},
  author = {Sören Christensen and Jan Kallsen and Claudia Strauch and Lukas Trottner},
  journal= {arXiv preprint arXiv:2510.23197},
  year   = {2025}
}
R2 v1 2026-07-01T07:07:29.314Z