English

Fast, robust approximate message passing

Data Structures and Algorithms 2024-11-06 v1 Machine Learning Machine Learning

Abstract

We give a fast, spectral procedure for implementing approximate-message passing (AMP) algorithms robustly. For any quadratic optimization problem over symmetric matrices XX with independent subgaussian entries, and any separable AMP algorithm A\mathcal A, our algorithm performs a spectral pre-processing step and then mildly modifies the iterates of A\mathcal A. If given the perturbed input X+ERn×nX + E \in \mathbb R^{n \times n} for any EE supported on a εn×εn\varepsilon n \times \varepsilon n principal minor, our algorithm outputs a solution v^\hat v which is guaranteed to be close to the output of A\mathcal A on the uncorrupted XX, with A(X)v^2f(ε)A(X)2\|\mathcal A(X) - \hat v\|_2 \le f(\varepsilon) \|\mathcal A(X)\|_2 where f(ε)0f(\varepsilon) \to 0 as ε0\varepsilon \to 0 depending only on ε\varepsilon.

Keywords

Cite

@article{arxiv.2411.02764,
  title  = {Fast, robust approximate message passing},
  author = {Misha Ivkov and Tselil Schramm},
  journal= {arXiv preprint arXiv:2411.02764},
  year   = {2024}
}

Comments

22 pages, 2 figures

R2 v1 2026-06-28T19:48:25.272Z