Unbiased Risk Estimates for Singular Value Thresholding and Spectral Estimators

DOI10.1109/TSP.2013.2270464zbMATH Open1393.94187arXiv1210.4139OpenAlexW2063942101MaRDI QIDQ4578740FDOQ4578740

Authors: Emmanuel J. Candès, Carlos Sing-Long, Joshua D. Trzasko 

Publication date: 22 August 2018

Published in: IEEE Transactions on Signal Processing (Search for Journal in Brave)

Abstract: In an increasing number of applications, it is of interest to recover an approximately low-rank data matrix from noisy observations. This paper develops an unbiased risk estimate---holding in a Gaussian model---for any spectral estimator obeying some mild regularity assumptions. In particular, we give an unbiased risk estimate formula for singular value thresholding (SVT), a popular estimation strategy which applies a soft-thresholding rule to the singular values of the noisy observations. Among other things, our formulas offer a principled and automated way of selecting regularization parameters in a variety of problems. In particular, we demonstrate the utility of the unbiased risk estimation for SVT-based denoising of real clinical cardiac MRI series data. We also give new results concerning the differentiability of certain matrix-valued functions.

Full work available at URL: https://arxiv.org/abs/1210.4139

Mathematics Subject Classification ID

Estimation in multivariate analysis (62H12) Signal theory (characterization, reconstruction, filtering, etc.) (94A12) Biomedical imaging and signal processing (92C55) Image processing (compression, reconstruction, etc.) in information and communication theory (94A08)

Cited In (41)

• Adaptive singular value shrinkage estimate for low rank tensor denoising
• Relaxing the Gaussian assumption in shrinkage and SURE in high dimension
• Selecting Regularization Parameters for Nuclear Norm--Type Minimization Problems
• A singular value thresholding with diagonal-update algorithm for low-rank matrix completion
• Ridge-type linear shrinkage estimation of the mean matrix of a high-dimensional normal distribution
• Empirical risk minimization as parameter choice rule for general linear regularization methods
• An adaptive singular value shrinkage for estimation problem of low-rank matrix mean with unknown covariance matrix
• Rate-optimal perturbation bounds for singular subspaces with applications to high-dimensional statistics
• Unbiased predictive risk estimation of the Tikhonov regularization parameter: convergence with increasing rank approximations of the singular value decomposition
• Excess Optimism: How Biased is the Apparent Error of an Estimator Tuned by SURE?
• The degrees of freedom of partly smooth regularizers
• Hierarchical nuclear norm penalization for multi-view data integration
• Accelerating patch-based low-rank image restoration using kd-forest and Lanczos approximation
• Speckle reduction in matrix-log domain for synthetic aperture radar imaging
• Low complexity regularization of linear inverse problems
• Adaptive shrinkage of singular values
• Degrees of freedom for off-the-grid sparse estimation
• Minimax risk of matrix denoising by singular value thresholding
• Unbiased risk estimates for matrix estimation in the elliptical case
• A rank-two relaxed parallel splitting version of the augmented Lagrangian method with step size in (0,2) for separable convex programming
• Regularised PCA to denoise and visualise data
• On Stein's unbiased risk estimate for reduced rank estimators
• Covariance matrix estimation under data-based loss
• Imputation and low-rank estimation with missing not at random data
• Empirical Bayes estimates for a two-way cross-classified model
• Second-order Stein: SURE for SURE and other applications in high-dimensional inference
• On Degrees of Freedom of Projection Estimators With Applications to Multivariate Nonparametric Regression
• Generalized SURE for optimal shrinkage of singular values in low-rank matrix denoising
• Generalized Factor Model for Ultra-High Dimensional Correlated Variables with Mixed Types
• Operator-Lipschitz estimates for the singular value functional calculus
• Risk estimators for choosing regularization parameters in ill-posed problems -- properties and limitations
• Computing the degrees of freedom of rank-regularized estimators and cousins
• Channel estimation for finite scatterers massive multi-user MIMO system
• Going Off the Grid: Iterative Model Selection for Biclustered Matrix Completion
• Proximal Markov chain Monte Carlo algorithms
• Matrix completion with nonconvex regularization: spectral operators and scalable algorithms
• Low-rank matrix denoising for count data using unbiased Kullback-Leibler risk estimation
• Heteroskedastic PCA: algorithm, optimality, and applications
• Degrees of freedom and model selection for \(k\)-means clustering
• Smooth singular value thresholding algorithm for low-rank matrix completion problem
• Scale matrix estimation of an elliptically symmetric distribution in high and low dimensions

This page was built for publication: Unbiased Risk Estimates for Singular Value Thresholding and Spectral Estimators

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4578740)