A multiscale cavity method for sublinear-rank symmetric matrix factorization
Mathematical Statistics and Learning · 18.03.2026 · 9 (1) · 1-68
DOI: 10.4171/msl/57 ↗Kurz gefasst
Fachartikel in der Zeitschrift Mathematical Statistics and Learning, erschienen am 18.03.2026. Verfasst von Jean Barbier und 2 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref).
Abstract (Originalsprache)
We consider a statistical model for symmetric matrix factorization with additive Gaussian noise in the high-dimensional regime, where the rank of the signal matrix to infer M scales with its size N as M=\mathrm{o}(\sqrt{\ln N}) . Allowing for an N -dependent rank offers new challenges and requires new methods. Working in the Bayes-optimal setting, we show that whenever the signal has i.i.d. entries, the limiting mutual information between signal and data is given by a variational formula involving a rank- one replica symmetric potential. In other words, from the information-theoretic perspective, the case of a (slowly) growing rank is the same as when M=1 (namely, the standard spiked Wigner model). The proof is primarily based on a novel multiscale cavity method allowing for growing rank along with some information-theoretic identities on worst noise for the vector Gaussian channel. We believe that the cavity method developed here will play a role in the analysis of a broader class of inference and spin models where the degrees of freedom are large arrays instead of vectors.
Zitieren
APA
Barbier, J., Ko, J., & Rahman, A. A. (2026). A multiscale cavity method for sublinear-rank symmetric matrix factorization. Mathematical Statistics and Learning, 9(1), 1-68. https://doi.org/10.4171/msl/57
Vancouver
Barbier J, Ko J, Rahman AA. A multiscale cavity method for sublinear-rank symmetric matrix factorization. Mathematical Statistics and Learning. 2026;9(1):1-68. doi:10.4171/msl/57
BibTeX
@article{Barbier2026,
author = {Barbier, Jean and Ko, Justin and Rahman, Anas A.},
title = {A multiscale cavity method for sublinear-rank symmetric matrix factorization},
journal = {Mathematical Statistics and Learning},
year = {2026},
volume = {9},
number = {1},
pages = {1--68},
doi = {10.4171/msl/57}
}
EUVIMED