Errors-in-variables models with dependent measurements

DOI10.1214/17-EJS1234MaRDI QIDQ527095zbMATH OpenFDO

Authors Shuheng Zhou, Mark Rudelson

Publication date 16 May 2017

Published in Electronic Journal of Statistics (Search for Journal in Brave)

Full work available at URL https://arxiv.org/abs/1611.04701, https://projecteuclid.org/euclid.ejs/1493107295

zbMATH Keywords

errors-in-variables modelmatrix variate distributionsmeasurement error datarandom matrixsubgaussian concentration

Mathematics Subject Classification ID

Random matrices (probabilistic aspects) (60B20) Asymptotic properties of parametric estimators (62F12) Linear regression; mixed models (62J05) Ridge regression; shrinkage estimators (Lasso) (62J07)

Abstract: Suppose that we observe $yinmathbb{R}^n$ and $Xinmathbb{R}^{nimesm}$ in the following errors-in-variables model: �egin{eqnarray*} y & = & X_0 �eta^* +epsilon \ X & = & X_0 + W, end{eqnarray*} where $X_0$ is an $nimesm$ design matrix with independent subgaussian row vectors, $epsiloninmathbb{R}^n$ is a noise vector and $W$ is a mean zero $nimesm$ random noise matrix with independent subgaussian column vectors, independent of $X_0$ and $epsilon$. This model is significantly different from those analyzed in the literature in the sense that we allow the measurement error for each covariate to be a dependent vector across its $n$ observations. Such error structures appear in the science literature when modeling the trial-to-trial fluctuations in response strength shared across a set of neurons. Under sparsity and restrictive eigenvalue type of conditions, we show that one is able to recover a sparse vector $$ from the model given a single observation matrix $X$ and the response vector $y$. We establish consistency in estimating $$ and obtain the rates of convergence in the $el{l}_q$ norm, where $q=1,2$. We show error bounds which approach that of the regular Lasso and the Dantzig selector in case the errors in $W$ are tending to 0. We analyze the convergence rates of the gradient descent methods for solving the nonconvex programs and show that the composite gradient descent algorithm is guaranteed to converge at a geometric rate to a neighborhood of the global minimizers: the size of the neighborhood is bounded by the statistical error in the $el{l}_2$ norm. Our analysis reveals interesting connections between computational and statistical efficiency and the concentration of measure phenomenon in random matrix theory. We provide simulation evidence illuminating the theoretical predictions.

Recommendations

• Linear and conic programming estimators in high dimensional errors-in-variables models
• A minimax approach to errors-in-variables linear models
• An \(\{\ell_{1},\ell_{2},\ell_{\infty}\}\)-regularization approach to high-dimensional errors-in-variables models
• Sparse estimation in high-dimensional linear errors-in-variables regression via a covariate relaxation method
• Estimation for the multivariate errors-in-variables model with estimated error covariance matrix

Cited in

(7)

• The set of observationally equivalent errors-in-variables models
• Robust inference in a linear functional model with replications using the \(t\) distribution
• Errors-in-Variables and Reverse Regression
• Regression from dependent observations
• scientific article; zbMATH DE number 2209978 (Why is no real title available?)
• Development of a first order integrated moving average model corrupted with a Markov modulated convex combination of autoregressive moving average errors
• Concentration of measure bounds for matrix-variate data with missing values

This page was built for publication: Errors-in-variables models with dependent measurements

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q527095)