Typical l₁-recovery limit of sparse vectors represented by concatenations of random orthogonal matrices

DOI10.1088/1742-5468/2012/12/P12003zbMATH Open1456.15036arXiv1208.4696OpenAlexW3101710166MaRDI QIDQ3301307FDOQ3301307

Yoshiyuki Kabashima, Saikat Chatterjee, Mikko Vehkaperä

Publication date: 11 August 2020

Published in: Journal of Statistical Mechanics: Theory and Experiment (Search for Journal in Brave)

Abstract: We consider the problem of recovering an $N$-dimensional sparse vector $vmx$ from its linear transformation $vmy=vmDvmx$ of $M(<N)$ dimension. Minimizing the $l_1$-norm of $vmx$ under the constraint $vmy=vmDvmx$ is a standard approach for the recovery problem, and earlier studies report that the critical condition for typically successful $l_1$-recovery is universal over a variety of randomly constructed matrices $vmD$. For examining the extent of the universality, we focus on the case in which $vmD$ is provided by concatenating $b=N/M$ matrices $vm{O}_1,vm{O}_2,\mathrm{...},vm{O}_b$ drawn uniformly according to the Haar measure on the $MimesM$ orthogonal matrices. By using the replica method in conjunction with the development of an integral formula for handling the random orthogonal matrices, we show that the concatenated matrices can result in better recovery performance than what the universality predicts when the density of non-zero signals is not uniform among the $b$ matrix modules. The universal condition is reproduced for the special case of uniform non-zero signal densities. Extensive numerical experiments support the theoretical predictions.

Full work available at URL: https://arxiv.org/abs/1208.4696

Recommendations

• Nonuniform sparse recovery with subgaussian matrices
• Sparsity and incoherence in orthogonal matching pursuit
• Simple bounds for recovering low-complexity models
• On exact recovery of sparse vectors from linear measurements
• Stable signal recovery from incomplete and inaccurate measurements

Mathematics Subject Classification ID

Signal theory (characterization, reconstruction, filtering, etc.) (94A12) Random matrices (algebraic aspects) (15B52)

Cites Work

• Optimally sparse representation in general (nonorthogonal) dictionaries via ℓ ¹ minimization
• Title not available (Why is that?)
• Sparse and redundant representations. From theory to applications in signal and image processing.
• Compressed sensing
• Uncertainty principles and ideal atomic decomposition
• The planar approximation. II
• High-dimensional centrally symmetric polytopes with neighborliness proportional to dimension
• Replica field theory for deterministic models. II. A non-random spin glass with glassy behaviour
• Counting faces of randomly projected polytopes when the projection radically lowers dimension
• Introduction to the replica theory of disordered statistical systems
• A generalized uncertainty principle and sparse representation in pairs of bases
• Perceptron capacity revisited: classification ability for correlated patterns

Cited In (5)

• Analysis of sparse recovery algorithms via the replica method
• Bayesian signal reconstruction for 1-bit compressed sensing
• \(L_2\)-norm sampling discretization and recovery of functions from RKHS with finite trace
• On the Fundamental Limits of Recovering Tree Sparse Vectors From Noisy Linear Measurements
• Replica analysis of overfitting in generalized linear regression models

This page was built for publication: Typical \(l_1\)-recovery limit of sparse vectors represented by concatenations of random orthogonal matrices

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3301307)