English

Spiked Covariance Estimation from Modulo-Reduced Measurements

Information Theory 2022-05-20 v3 math.IT Statistics Theory Machine Learning Statistics Theory

Abstract

Consider the rank-1 spiked model: X=νξu+Z\bf{X}=\sqrt{\nu}\xi \bf{u}+ \bf{Z}, where ν\nu is the spike intensity, uSk1\bf{u}\in\mathbb{S}^{k-1} is an unknown direction and ξN(0,1),ZN(0,I)\xi\sim \mathcal{N}(0,1),\bf{Z}\sim \mathcal{N}(\bf{0},\bf{I}). Motivated by recent advances in analog-to-digital conversion, we study the problem of recovering uSk1\bf{u}\in \mathbb{S}^{k-1} from nn i.i.d. modulo-reduced measurements Y=[X]modΔ\bf{Y}=[\bf{X}]\mod \Delta, focusing on the high-dimensional regime (k1k\gg 1). We develop and analyze an algorithm that, for most directions u\bf{u} and ν=poly(k)\nu=\mathrm{poly}(k), estimates u\bf{u} to high accuracy using n=poly(k)n=\mathrm{poly}(k) measurements, provided that Δlogk\Delta\gtrsim \sqrt{\log k}. Up to constants, our algorithm accurately estimates u\bf{u} at the smallest possible Δ\Delta that allows (in an information-theoretic sense) to recover X\bf{X} from Y\bf{Y}. A key step in our analysis involves estimating the probability that a line segment of length ν\approx\sqrt{\nu} in a random direction u\bf{u} passes near a point in the lattice ΔZk\Delta \mathbb{Z}^k. Numerical experiments show that the developed algorithm performs well even in a non-asymptotic setting.

Keywords

Cite

@article{arxiv.2110.01150,
  title  = {Spiked Covariance Estimation from Modulo-Reduced Measurements},
  author = {Elad Romanov and Or Ordentlich},
  journal= {arXiv preprint arXiv:2110.01150},
  year   = {2022}
}

Comments

AISTATS, 2022

R2 v1 2026-06-24T06:35:34.664Z