Restricted Eigenvalue Condition Research Articles

Group Logistic Regression Models with lp,q Regularization

In this paper, we proposed a logistic regression model with lp,q regularization that could give a group sparse solution. The model could be applied to variable-selection problems with sparse group structures. In the context of big data, the solutions for practical problems are often group sparse, so it is necessary to study this kind of model. We defined the model from three perspectives: theoretical, algorithmic and numeric. From the theoretical perspective, by introducing the notion of the group restricted eigenvalue condition, we gave the oracle inequality, which was an important property for the variable-selection problems. The global recovery bound was also established for the logistic regression model with lp,q regularization. From the algorithmic perspective, we applied the well-known alternating direction method of multipliers (ADMM) algorithm to solve the model. The subproblems for the ADMM algorithm were solved effectively. From the numerical perspective, we performed experiments for simulated data and real data in the factor stock selection. We employed the ADMM algorithm that we presented in the paper to solve the model. The numerical results were also presented. We found that the model was effective in terms of variable selection and prediction.

Moving beyond sub-Gaussianity in high-dimensional statistics: applications in covariance estimation and linear regression

Abstract Concentration inequalities form an essential toolkit in the study of high-dimensional statistical methods. Most of the relevant statistics literature in this regard is, however, based on the assumptions of sub-Gaussian or sub-exponential random variables/vectors. In this paper, we first bring together, through a unified exposition, various probabilistic inequalities for sums of independent random variables under much more general exponential type (namely sub-Weibull) tail assumptions. These results extract a part sub-Gaussian tail behavior of the sum in finite samples, matching the asymptotics governed by the central limit theorem, and are compactly represented in terms of a new Orlicz quasi-norm—the Generalized Bernstein–Orlicz norm—that typifies such kind of tail behaviors. We illustrate the usefulness of these inequalities through the analysis of four fundamental problems in high-dimensional statistics. In the first two problems, we study the rate of convergence of the sample covariance matrix in terms of the maximum elementwise norm and the maximum $k$-sub-matrix operator norm that are key quantities of interest in bootstrap procedures and high-dimensional structured covariance matrix estimation, as well as in high-dimensional and post-selection inference. The third example concerns the restricted eigenvalue condition, required in high-dimensional linear regression, which we verify for all sub-Weibull random vectors through a unified analysis, and also prove a more general result related to restricted strong convexity in the process. In the final example, we consider the Lasso estimator for linear regression and establish its rate of convergence to be generally $\sqrt{k\log p/n}$, for $k$-sparse signals, under much weaker than usual tail assumptions (on the errors as well as the covariates), while also allowing for misspecified models and both fixed and random design. To our knowledge, these are the first such results for Lasso obtained in this generality. The common feature in all our results over all the examples is that the convergence rates under most exponential tails match the usual (optimal) ones obtained under sub-Gaussian assumptions. Finally, we also establish some complementary results on analogous tail bounds for the suprema of empirical processes indexed by sub-Weibull variables. All our results are finite samples.

Calibrated zero-norm regularized LS estimator for high-dimensional error-in-variables regression

This paper is concerned with high-dimensional error-in-variables regression that aims at identifying a small number of important interpretable factors for corrupted data from many applications where measurement errors or missing data can not be ignored. Motivated by CoCoLasso due to Datta and Zou \cite{Datta16} and the advantage of the zero-norm regularized LS estimator over Lasso for clean data, we propose a calibrated zero-norm regularized LS (CaZnRLS) estimator by constructing a calibrated least squares loss with a positive definite projection of an unbiased surrogate for the covariance matrix of covariates, and use the multi-stage convex relaxation approach to compute the CaZnRLS estimator. Under a restricted eigenvalue condition on the true matrix of covariates, we derive the $\ell_2$-error bound of every iterate and establish the decreasing of the error bound sequence, and the sign consistency of the iterates after finite steps. The statistical guarantees are also provided for the CaZnRLS estimator under two types of measurement errors. Numerical comparisons with CoCoLasso and NCL (the nonconvex Lasso proposed by Poh and Wainwright \cite{Loh11}) demonstrate that CaZnRLS not only has the comparable or even better relative RSME but also has the least number of incorrect predictors identified.

A multi-stage convex relaxation approach to noisy structured low-rank matrix recovery

This paper concerns with a noisy structured low-rank matrix recovery problem which can be modeled as a structured rank minimization problem. We reformulate this problem as a mathematical program with a generalized complementarity constraint (MPGCC), and show that its penalty version, yielded by moving the generalized complementarity constraint to the objective, has the same global optimal solution set as the MPGCC does whenever the penalty parameter is over a threshold. Then, by solving the exact penalty problem in an alternating way, we obtain a multi-stage convex relaxation approach. We provide theoretical guarantees for our approach under a mild restricted eigenvalue condition, by quantifying the reduction of the error and approximate rank bounds of the first stage convex relaxation (which is exactly the nuclear norm relaxation) in the subsequent stages and establishing the geometric convergence of the error sequence in a statistical sense. Numerical experiments are conducted for some structured low-rank matrix recovery examples to confirm our theoretical findings.

Graph-based regularization for regression problems with alignment and highly-correlated designs.

Sparse models for high-dimensional linear regression and machine learning have received substantial attention over the past two decades. Model selection, or determining which features or covariates are the best explanatory variables, is critical to the interpretability of a learned model. Much of the current literature assumes that covariates are only mildly correlated. However, in many modern applications covariates are highly correlated and do not exhibit key properties (such as the restricted eigenvalue condition, restricted isometry property, or other related assumptions). This work considers a high-dimensional regression setting in which a graph governs both correlations among the covariates and the similarity among regression coefficients - meaning there is alignment between the covariates and regression coefficients. Using side information about the strength of correlations among features, we form a graph with edge weights corresponding to pairwise covariances. This graph is used to define a graph total variation regularizer that promotes similar weights for correlated features. This work shows how the proposed graph-based regularization yields mean-squared error guarantees for a broad range of covariance graph structures. These guarantees are optimal for many specific covariance graphs, including block and lattice graphs. Our proposed approach outperforms other methods for highly-correlated design in a variety of experiments on synthetic data and real biochemistry data.

Sorted concave penalized regression

The Lasso is biased. Concave penalized least squares estimation (PLSE) takes advantage of signal strength to reduce this bias, leading to sharper error bounds in prediction, coefficient estimation and variable selection. For prediction and estimation, the bias of the Lasso can be also reduced by taking a smaller penalty level than what selection consistency requires, but such smaller penalty level depends on the sparsity of the true coefficient vector. The sorted $\ell_{1}$ penalized estimation (Slope) was proposed for adaptation to such smaller penalty levels. However, the advantages of concave PLSE and Slope do not subsume each other. We propose sorted concave penalized estimation to combine the advantages of concave and sorted penalizations. We prove that sorted concave penalties adaptively choose the smaller penalty level and at the same time benefits from signal strength, especially when a significant proportion of signals are stronger than the corresponding adaptively selected penalty levels. A local convex approximation for sorted concave penalties, which extends the local linear and quadratic approximations for separable concave penalties, is developed to facilitate the computation of sorted concave PLSE and proven to possess desired prediction and estimation error bounds. Our analysis of prediction and estimation errors requires the restricted eigenvalue condition on the design, not beyond, and provides selection consistency under a required minimum signal strength condition in addition. Thus, our results also sharpens existing results on concave PLSE by removing the upper sparse eigenvalue component of the sparse Riesz condition.

High-Dimensional Classification by Sparse Logistic Regression

We consider high-dimensional binary classification by sparse logistic regression. We propose a model/feature selection procedure based on penalized maximum likelihood with a complexity penalty on the model size and derive the non-asymptotic bounds for its misclassification excess risk. To assess its tightness, we establish the corresponding minimax lower bounds. The bounds can be reduced under the additional low-noise condition. The proposed complexity penalty is remarkably related to the Vapnik-Chervonenkis-dimension of a set of sparse linear classifiers. Implementation of any complexity penalty-based criterion, however, requires a combinatorial search over all possible models. To find a model selection procedure computationally feasible for high-dimensional data, we extend the Slope estimator for logistic regression and show that under an additional weighted restricted eigenvalue condition it is rate-optimal in the minimax sense.

Slope meets Lasso: Improved oracle bounds and optimality

We show that two polynomial time methods, a Lasso estimator with adaptively chosen tuning parameter and a Slope estimator, adaptively achieve the minimax prediction and $\ell_{2}$ estimation rate $(s/n)\log(p/s)$ in high-dimensional linear regression on the class of $s$-sparse vectors in $\mathbb{R}^{p}$. This is done under the Restricted Eigenvalue (RE) condition for the Lasso and under a slightly more constraining assumption on the design for the Slope. The main results have the form of sharp oracle inequalities accounting for the model misspecification error. The minimax optimal bounds are also obtained for the $\ell_{q}$ estimation errors with $1\le q\le2$ when the model is well specified. The results are nonasymptotic, and hold both in probability and in expectation. The assumptions that we impose on the design are satisfied with high probability for a large class of random matrices with independent and possibly anisotropically distributed rows. We give a comparative analysis of conditions, under which oracle bounds for the Lasso and Slope estimators can be obtained. In particular, we show that several known conditions, such as the RE condition and the sparse eigenvalue condition are equivalent if the $\ell_{2}$-norms of regressors are uniformly bounded.

Integrative multi-view regression: Bridging group-sparse and low-rank models.

Multi-view data have been routinely collected in various fields of science and engineering. A general problem is to study the predictive association between multivariate responses and multi-view predictor sets, all of which can be of high dimensionality. It is likely that only a few views are relevant to prediction, and the predictors within each relevant view contribute to the prediction collectively rather than sparsely. We cast this new problem under the familiar multivariate regression framework and propose an integrative reduced-rank regression (iRRR), where each view has its own low-rank coefficient matrix. As such, latent features are extracted from each view in a supervised fashion. For model estimation, we develop a convex composite nuclear norm penalization approach, which admits an efficient algorithm via alternating direction method of multipliers. Extensions to non-Gaussian and incomplete data are discussed. Theoretically, we derive non-asymptotic oracle bounds of iRRR under a restricted eigenvalue condition. Our results recover oracle bounds of several special cases of iRRR including Lasso, group Lasso, and nuclear norm penalized regression. Therefore, iRRR seamlessly bridges group-sparse and low-rank methods and can achieve substantially faster convergence rate under realistic settings of multi-view learning. Simulation studies and an application in the Longitudinal Studies of Aging further showcase the efficacy of the proposed methods.

Minimax Optimal Convex Methods for Poisson Inverse Problems Under <inline-formula> <tex-math notation="LaTeX">$\ell_{q}$ </tex-math> </inline-formula>-Ball Sparsity

In this paper, we study the minimax rates and provide an implementable convex algorithm for Poisson inverse problems under weak sparsity and physical constraints. In particular, we assume the model $y_{i} \sim \mathrm {Poisson}(Ta_{i}^{\top }f^{*})$ for $1 \leq i \leq n$ , where $T \in \mathbb {R}_{+}$ is the intensity, and we impose weak sparsity on $f^{*} \in \mathbb {R}^{p}$ by assuming $f^{*}$ lies in an $\ell _{q}$ -ball when rotated according to an orthonormal basis $D \in \mathbb {R}^{p \times p}$ . In addition, since we are modeling real-physical systems, we also impose positivity and flux-preserving constraints on the matrix $A = [a_{1}, a_{2},\ldots, a_{n}]^{\top }$ and the function $f^{*}$ . We prove minimax lower bounds for this model, which scale as $R_{q} ({\log p}/{T})^{1 - ({q}/{2})}$ where it is noticeable that the rate depends on the intensity $T$ and not the sample size $n$ . We also show that an $\ell _{1}$ -based regularized least-squares estimator achieves this minimax lower bound, provided a suitable restricted eigenvalue condition is satisfied. Finally, we prove that provided $n \geq \tilde {K} \log p$ where $\tilde {K} = \Theta (R_{q} ({\log p}/{T})^{- ({q}/{2})})$ represents an approximate sparsity level, and our restricted eigenvalue condition and physical constraints are satisfied for random bounded ensembles. We also provide numerical experiments that validate our mean-squared error bounds. Our results address a number of open issues from prior work on Poisson inverse problems that focuses on strictly sparse models and does not provide guarantees for convex implementable algorithms.

Signal recovery under cumulative coherence

This paper considers signal recovery in the framework of cumulative coherence. First, we show that the Lasso estimator and the Dantzig selector exhibit similar behavior under the cumulative coherence. Then we estimate the approximation equivalence between the Lasso and the Dantzig selector by calculating prediction loss difference under the condition of cumulative coherence. And we also prove that the cumulative coherence implies the restricted eigenvalue condition. Last, we illustrate the advantages of cumulative coherence condition for three class matrices, in terms of the recovery performance of sparse signals via extensive numerical experiments.

Oracle inequalities for sparse additive quantile regression in reproducing kernel Hilbert space

This paper considers the estimation of the sparse additive quantile regression (SAQR) in high-dimensional settings. Given the nonsmooth nature of the quantile loss function and the nonparametric complexities of the component function estimation, it is challenging to analyze the theoretical properties of ultrahigh-dimensional SAQR. We propose a regularized learning approach with a two-fold Lasso-type regularization in a reproducing kernel Hilbert space (RKHS) for SAQR. We establish nonasymptotic oracle inequalities for the excess risk of the proposed estimator without any coherent conditions. If additional assumptions including an extension of the restricted eigenvalue condition are satisfied, the proposed method enjoys sharp oracle rates without the light tail requirement. In particular, the proposed estimator achieves the minimax lower bounds established for sparse additive mean regression. As a by-product, we also establish the concentration inequality for estimating the population mean when the general Lipschitz loss is involved. The practical effectiveness of the new method is demonstrated by competitive numerical results.

Two-stage convex relaxation approach to low-rank and sparsity regularized least squares loss

In this paper we consider the rank and zero norm regularized least squares loss minimization problem with a spectral norm ball constraint. For this class of NP-hard optimization problems, we propose a two-stage convex relaxation approach by majorizing some suitable locally Lipschitz continuous surrogates. Furthermore, the Frobenius norm error bound for the optimal solution of each stage is characterized and the theoretical guarantee is established for the two-stage convex relaxation approach by showing that the error bound of the first stage convex relaxation (i.e., the nuclear norm and $$\ell _1$$ -norm regularized minimization problem), can be reduced much by the second stage convex relaxation under a suitable restricted eigenvalue condition. Also, we verify the efficiency of the proposed approach by applying it to some random test problems and some real problems.

Sparse recovery under weak moment assumptions

We prove that iid random vectors that satisfy a rather weak moment assumption can be used as measurement vectors in Compressed Sensing, and the number of measurements required for exact reconstruction is the same as the best possible estimate – exhibited by a random Gaussian matrix. We also prove that this moment condition is necessary, up to a log log factor. In addition, we explore the Compatibility Condition and the Restricted Eigenvalue Condition in the noisy setup, as well as properties of neighbourly random polytopes.

Folded concave penalized sparse linear regression: sparsity, statistical performance, and algorithmic theory for local solutions.

This paper concerns the folded concave penalized sparse linear regression (FCPSLR), a class of popular sparse recovery methods. Although FCPSLR yields desirable recovery performance when solved globally, computing a global solution is NP-complete. Despite some existing statistical performance analyses on local minimizers or on specific FCPSLR-based learning algorithms, it still remains open questions whether local solutions that are known to admit fully polynomial-time approximation schemes (FPTAS) may already be sufficient to ensure the statistical performance, and whether that statistical performance can be non-contingent on the specific designs of computing procedures. To address the questions, this paper presents the following threefold results: (i) Any local solution (stationary point) is a sparse estimator, under some conditions on the parameters of the folded concave penalties. (ii) Perhaps more importantly, any local solution satisfying a significant subspace second-order necessary condition (S3ONC), which is weaker than the second-order KKT condition, yields a bounded error in approximating the true parameter with high probability. In addition, if the minimal signal strength is sufficient, the S3ONC solution likely recovers the oracle solution. This result also explicates that the goal of improving the statistical performance is consistent with the optimization criteria of minimizing the suboptimality gap in solving the non-convex programming formulation of FCPSLR. (iii) We apply (ii) to the special case of FCPSLR with minimax concave penalty (MCP) and show that under the restricted eigenvalue condition, any S3ONC solution with a better objective value than the Lasso solution entails the strong oracle property. In addition, such a solution generates a model error (ME) comparable to the optimal but exponential-time sparse estimator given a sufficient sample size, while the worst-case ME is comparable to the Lasso in general. Furthermore, to guarantee the S3ONC admits FPTAS.

Analysis Recovery With Coherent Frames and Correlated Measurements

This paper introduces the restricted eigenvalue condition adapted to frame ${D}$ ( ${D}$ -RE), which is a natural extension to the standard restricted eigenvalue condition. The ${D}$ -RE condition is a relaxation of the ${D^\dagger }$ -RIP, where ${D^\dagger =(DD^{*})^{-1}D}$ is the canonical dual frame of ${D}$ . We establish the ${D}$ -RE condition for several classes of correlated measurement matrices, when the covariance matrix of row measurements satisfies the ${D}$ -RE condition. Furthermore, by the ${D}$ -RE condition, we get the error bounds in the analysis LASSO (ALASSO) and the analysis Dantzig Selector (ADS) under a sparsity scenario. In order to recover non-sparse signals, we consider the robust ${\ell _{2}}~{D}$ -nullspace property of correlated Gaussian matrices. Similarly, we get the error estimations in the ALASSO and the ADS in non-sparse case. The approximation equivalence between the ALASSO and the ADS is also established by calculating prediction loss difference.

An analysis of penalized interaction models

An important consideration for variable selection in interaction models is to design an appropriate penalty that respects hierarchy of the importance of the variables. A common theme is to include an interaction term only after the corresponding main effects are present. In this paper, we study several recently proposed approaches and present a unified analysis on the convergence rate for a class of estimators, when the design satisfies the restricted eigenvalue condition. In particular, we show that with probability tending to one, the resulting estimates have a rate of convergence $s\sqrt{\log p_1/n}$ in the $\ell_1$ error, where $p_1$ is the ambient dimension, $s$ is the true dimension and $n$ is the sample size. We give a new proof that the restricted eigenvalue condition holds with high probability, when the variables in the main effects and the errors follow sub-Gaussian distributions. Under this setup, the interactions no longer follow Gaussian or sub-Gaussian distributions even if the main effects follow Gaussian, and thus existing works are not applicable. This result is of independent interest.

Two-stage convex relaxation approach to least squares loss constrained low-rank plus sparsity optimization problems

This paper is concerned with the least squares loss constrained low-rank plus sparsity optimization problems that seek a low-rank matrix and a sparse matrix by minimizing a positive combination of the rank function and the zero norm over a least squares constraint set describing the observation or prior information on the target matrix pair. For this class of NP-hard optimization problems, we propose a two-stage convex relaxation approach by the majorization for suitable locally Lipschitz continuous surrogates, which have a remarkable advantage in reducing the error yielded by the popular nuclear norm plus $$\ell _1$$l1-norm convex relaxation method. Also, under a suitable restricted eigenvalue condition, we establish a Frobenius norm error bound for the optimal solution of each stage and show that the error bound of the first stage convex relaxation (i.e. the nuclear norm plus $$\ell _1$$l1-norm convex relaxation), can be reduced much by the second stage convex relaxation, thereby providing the theoretical guarantee for the two-stage convex relaxation approach. We also verify the efficiency of the proposed approach by applying it to some random test problems and some problems with real data arising from specularity removal from face images, and foreground/background separation from surveillance videos.

Robustness of Sparse Recovery via <inline-formula> <tex-math notation="LaTeX">$F$ </tex-math></inline-formula>-Minimization: A Topological Viewpoint

A recent trend in compressed sensing is to consider non-convex optimization techniques for sparse recovery. The important case of $F$-minimization has become of particular interest, for which the exact reconstruction condition (ERC) in the noiseless setting can be precisely characterized by the null space property (NSP). However, little work has been done concerning its robust reconstruction condition (RRC) in the noisy setting. We look at the null space of the measurement matrix as a point on the Grassmann manifold, and then study the relation between the ERC and RRC sets, denoted as $\Omega_J$ and $\Omega_J^r$, respectively. It is shown that $\Omega_J^r$ is the interior of $\Omega_J$, from which a previous result of the equivalence of ERC and RRC for $\ell_p$-minimization follows easily as a special case. Moreover, when $F$ is non-decreasing, it is shown that $\overline{\Omega}_J\setminus\interior(\Omega_J)$ is a set of measure zero and of the first category. As a consequence, the probabilities of ERC and RRC are the same if the measurement matrix $\mathbf{A}$ is randomly generated according to a continuous distribution. Quantitatively, if the null space $\mathcal{N}(\bf A)$ lies in the "$d$-interior" of $\Omega_J$, then RRC will be satisfied with the robustness constant $C=\frac{2+2d}{d\sigma_{\min}(\mathbf{A}^{\top})}$; and conversely if RRC holds with $C=\frac{2-2d}{d\sigma_{\max}(\mathbf{A}^{\top})}$, then $\mathcal{N}(\bf A)$ must lie in $d$-interior of $\Omega_J$. We also present several rules for comparing the performances of different cost functions. Finally, these results are capitalized to derive achievable tradeoffs between the measurement rate and robustness with the aid of Gordon's escape through the mesh theorem or a connection between NSP and the restricted eigenvalue condition.

Nystrom Approximation for Sparse Kernel Methods: Theoretical Analysis and Empirical Evaluation

Nystrom approximation is an effective approach to accelerate the computation of kernel matrices in many kernel methods. In this paper, we consider the Nystrom approximation for sparse kernel methods. Instead of relying on the low-rank assumption of the original kernels, which sometimes does not hold in some applications, we take advantage of the restricted eigenvalue condition, which has been proved to be robust for sparse kernel methods. Based on the restricted eigenvalue condition, we have provided not only the approximation bound for the original kernel matrix but also the recovery bound for the sparse solutions of sparse kernel regression. In addition to the theoretical analysis, we also demonstrate the good performance of the Nystrom approximation for sparse kernel regression on real world data sets.