Asymptotics and concentration bounds for bilinear forms of spectral projectors of sample covariance

DOI10.1214/15-AIHP705zbMATH Open1353.62053arXiv1408.4643OpenAlexW2963770579MaRDI QIDQ503099FDOQ503099

Authors: K. Lounici, Vladimir Koltchinskii 

Publication date: 11 January 2017

Published in: Annales de l'Institut Henri Poincaré. Probabilités et Statistiques (Search for Journal in Brave)

Abstract: Let $X,X_1,dots,X_n$ be i.i.d. Gaussian random variables with zero mean and covariance operator $Sigma=mathbbE(XotimesX)$ taking values in a separable Hilbert space $mathbbH.$ Let ${\displaystyle }$ {�f r}(Sigma):=frac{{ m tr}(Sigma)}{|Sigma|_{infty}} ${\displaystyle }$ be the effective rank of $Sigma,$ $mtr(Sigma)$ being the trace of $Sigma$ and $|Sigma{|}_{infty}$ being its operator norm. Let ${\displaystyle }$hat Sigma_n:=n^{-1}sum_{j=1}^n (X_jotimes X_j)${\displaystyle }$ be the sample (empirical) covariance operator based on $(X_1,dots,X_n).$ The paper deals with a problem of estimation of spectral projectors of the covariance operator $Sigma$ by their empirical counterparts, the spectral projectors of $hatSigm{a}_n$ (empirical spectral projectors). The focus is on the problems where both the sample size $n$ and the effective rank $$ are large. This framework includes and generalizes well known high-dimensional spiked covariance models. Given a spectral projector $P_r$ corresponding to an eigenvalue $m{u}_r$ of covariance operator $Sigma$ and its empirical counterpart $hat{P}_r,$ we derive sharp concentration bounds for bilinear forms of empirical spectral projector $hat{P}_r$ in terms of sample size $n$ and effective dimension $$ Building upon these concentration bounds, we prove the asymptotic normality of bilinear forms of random operators $hat{P}_r-mathbbEhat{P}_r$ under the assumptions that $noinfty$ and $$ In a special case of eigenvalues of multiplicity one, these results are rephrased as concentration bounds and asymptotic normality for linear forms of empirical eigenvectors. Other results include bounds on the bias $mathbbEhat{P}_r-P_r$ and a method of bias reduction as well as a discussion of possible applications to statistical inference in high-dimensional principal component analysis.

Full work available at URL: https://arxiv.org/abs/1408.4643

Recommendations

• Normal approximation and concentration of spectral projectors of sample covariance
• Bootstrap confidence sets for spectral projectors of sample covariance
• Asymptotically efficient estimation of smooth functionals of covariance operators
• Relative perturbation bounds with applications to empirical covariance operators
• Asymptotic efficiency in high-dimensional covariance estimation

zbMATH Keywords

principal component analysisasymptotic distributionperturbation theoryconcentration inequalitiessample covariancespectral projectorseffective rank

Mathematics Subject Classification ID

Nonparametric regression and quantile regression (62G08) Asymptotic properties of nonparametric inference (62G20) Factor analysis and principal components; correspondence analysis (62H25) Estimation in multivariate analysis (62H12) Random matrices (probabilistic aspects) (60B20)

Cited In (35)

• Title not available (Why is that?)
• Relative perturbation bounds with applications to empirical covariance operators
• A survey of high dimension low sample size asymptotics
• The Sup-norm Perturbation of HOSVD and Low Rank Tensor Denoising
• Robust dimension-free Gram operator estimates
• Entrywise eigenvector analysis of random matrices with low expected rank
• Normal approximation and confidence region of singular subspaces
• Non-asymptotic properties of spectral decomposition of large Gram-type matrices and applications
• Functional principal subspace sampling for large scale functional data analysis
• Wald Statistics in high-dimensional PCA
• On some extensions of Bernstein's inequality for self-adjoint operators
• Efficient estimation of smooth functionals in Gaussian shift models
• Perturbation of linear forms of singular vectors under Gaussian noise
• Quantitative limit theorems and bootstrap approximations for empirical spectral projectors
• Asymptotically efficient estimation of smooth functionals of covariance operators
• Subspace estimation from unbalanced and incomplete data matrices: \({\ell_{2,\infty}}\) statistical guarantees
• New asymptotic results in principal component analysis
• Efficient estimation of linear functionals of principal components
• Convergence rate of Krasulina estimator
• Bayesian inference for spectral projectors of the covariance matrix
• Distributed estimation of principal eigenspaces
• Spectral method and regularized MLE are both optimal for top-\(K\) ranking
• Estimation of smooth functionals in normal models: bias reduction and asymptotic efficiency
• Optimality of spectral clustering in the Gaussian mixture model
• Perturbation bounds for eigenspaces under a relative gap condition
• Principal component analysis in the local differential privacy model
• Leave-one-out singular subspace perturbation analysis for spectral clustering
• Concentration inequalities and moment bounds for sample covariance operators
• Nonasymptotic upper bounds for the reconstruction error of PCA
• Randomized estimation of functional covariance operator via subsampling
• Multifidelity Dimension Reduction via Active Subspaces
• Singular vector and singular subspace distribution for the matrix denoising model
• Asymptotic Theory of Eigenvectors for Random Matrices With Diverging Spikes
• An \({\ell_p}\) theory of PCA and spectral clustering
• Normal approximation and concentration of spectral projectors of sample covariance

This page was built for publication: Asymptotics and concentration bounds for bilinear forms of spectral projectors of sample covariance

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q503099)