Information limits and Thouless-Anderson-Palmer equations for spiked matrix models with structured noise

TL;DR

This paper characterizes the fundamental limits of Bayesian inference in structured spiked matrix models with general noise, revealing equivalences and proposing an efficient algorithm based on TAP equations.

Contribution

It provides the first information-theoretic limit characterization for models with general trace ensemble noise and introduces an algorithm inspired by TAP equations to achieve these limits.

Findings

Established the asymptotic equivalence between structured noise models and Gaussian models.

Derived the first characterization of information-theoretic limits for general trace ensemble noise.

Proposed an efficient algorithm that saturates the theoretical statistical limits.

Abstract

We consider a prototypical problem of Bayesian inference for a structured spiked model: a low-rank signal is corrupted by additive noise. While both information-theoretic and algorithmic limits are well understood when the noise is a Gaussian Wigner matrix, the more realistic case of structured noise still proves to be challenging. To capture the structure while maintaining mathematical tractability, a line of work has focused on rotationally invariant noise. However, existing studies either provide sub-optimal algorithms or are limited to special cases of noise ensembles. In this paper, using tools from statistical physics (replica method) and random matrix theory (generalized spherical integrals) we establish the first characterization of the information-theoretic limits for a noise matrix drawn from a general trace ensemble. Remarkably, our analysis unveils the asymptotic equivalence between the rotationally invariant model and a surrogate Gaussian one. Finally, we show how to saturate the predicted statistical limits using an efficient algorithm inspired by the theory of adaptive Thouless-Anderson-Palmer (TAP) equations.