Approximation properties of certain operator-induced norms on Hilbert spaces

DOI10.1016/J.JAT.2011.11.002MaRDI QIDQ765689zbMATH OpenOpenAlexFDO

Authors Martin J. Wainwright, Arash A. Amini

Publication date 22 March 2012

Published in Journal of Approximation Theory (Search for Journal in Brave)

Full work available at URL https://arxiv.org/abs/1105.6351

zbMATH Keywords

\(L^2\)-approximationanalysis of \(M\)-estimatorsempirical normHilbert spaces with reproducing kernelsSobolov functionals

Mathematics Subject Classification ID

Miscellaneous inequalities involving matrices (15A45) Approximation by arbitrary nonlinear expressions; widths and entropy (41A46) Orthogonal functions and polynomials, general theory of nontrigonometric harmonic analysis (42C05) Hilbert and pre-Hilbert spaces: geometry and topology (including spaces with semidefinite inner product) (46C05) Sobolev spaces and other spaces of ``smooth functions, embedding theorems, trace theorems (46E35) Linear operators in reproducing-kernel Hilbert spaces (including de Branges, de Branges-Rovnyak, and other structured spaces) (47B32) Asymptotic properties of parametric estimators (62F12)

Abstract: We consider a class of operator-induced norms, acting as finite-dimensional surrogates to the L2 norm, and study their approximation properties over Hilbert subspaces of L2 . The class includes, as a special case, the usual empirical norm encountered, for example, in the context of nonparametric regression in reproducing kernel Hilbert spaces (RKHS). Our results have implications to the analysis of M-estimators in models based on finite-dimensional linear approximation of functions, and also to some related packing problems.

Review text

Let \(L^2\) be the space of square-integrable functions on a compact set in \(\mathbb{R}^d\) with regard to a probabilistic measure, and let \(\|\cdot\|_{L^2}\) be the corresponding norm. If \(\mathcal{H}\) is a Hilbert space, \(\mathcal{H}\subset L^2\) and \(\Phi\) is a vector \(\Phi=(\Phi_1,\dotsc,\Phi_n)\) of continuous linear functionals \(\Phi_i:\mathcal{H}\to\mathbb{R}\), then one can define a norm on \(\mathcal{H}\) by \(\| f\|_{\mathcal{H}}=\sqrt{\sum_{i=1}^n(\Phi_i(f))^2}\), \(f\in\mathcal{H}\). This norm, a substitute of the norm \(\|\cdot\|_{L^2}\), is very useful for practical reasons. The authors are interested in the problem of determining the order of deviation between the two norms. More exactly, the authors consider the space \(\mathcal{H}=\{f\in L^2\mid f=\sum_{k=1}^{\infty}\sqrt{\sigma_k}\alpha_k\Psi_k, \text{ for some }(\alpha_k)_{k=1}^{\infty}\in l_2\}\), where \((\Psi_k)_{k=1}^{\infty}\) is an orthogonal sequence in \(L^2\) and \(\sigma_1\geq\sigma_2\geq\dotsb>0\) is a sequence which decreases to zero. The inner product is given by \(\langle f,g\rangle_{\mathcal{H}}=\sum_{k=1}^{\infty}\alpha_k\beta_k\), if \(f=\sum_{k=1}^{\infty}\sqrt{\sigma_k}\alpha_k\Psi_k\) and \(g=\sum_{k=1}^{\infty}\sqrt{\sigma_k}\beta_k\Psi_k\). The main result gives an upper bound for the quantity \(R_{\Phi}(\varepsilon)=\sup\{\| f\|_{L^2}^2\,|\,f\in\mathcal{H},\;\| f\|_{\mathcal{H}}^2\leq\varepsilon^2\}\). Then, several applications are discussed, namely, reproducing kernel Hilbert space, random domain sampling, Sobolov kernel and Fourier truncation.

Recommendations

• Approximation of linear operators in the 2-norm
• Kernel approximation on manifolds. II: The \(L_{\infty}\) norm of the \(L_2\) projector
• Learning theory estimates via integral operators and their approximations
• Approximating functions in reproducing kernel Hilbert spaces via statistical learning theory
• Some properties of Gaussian reproducing kernel Hilbert spaces and their implications for function approximation and learning theory

Cites work

• A tail inequality for suprema of unbounded empirical processes with applications to Markov chains
• Analysis II. Convex analysis and approximation theory. Transl. from the Russian by D. Newton
• Derandomizing the Ahlswede-Winter matrix-valued Chernoff bound using pessimistic estimators, and applications
• scientific article; zbMATH DE number 2089352 (Why is no real title available?)
• scientific article; zbMATH DE number 3877692 (Why is no real title available?)
• scientific article; zbMATH DE number 5654889 (Why is no real title available?)
• scientific article; zbMATH DE number 3966959 (Why is no real title available?)
• scientific article; zbMATH DE number 50713 (Why is no real title available?)
• Inequalities. A journey into linear analysis
• Invertibility of symmetric random matrices
• Local Rademacher complexities
• Local Rademacher complexities and oracle inequalities in risk minimization. (2004 IMS Medallion Lecture). (With discussions and rejoinder)
• Minimax-optimal rates for sparse additive models over kernel classes via convex programming
• Random vectors in the isotropic position
• Strong converse for identification via quantum channels
• The concentration of measure phenomenon
• Weak convergence and empirical processes. With applications to statistics

Cited in

(4)

• On the \(L_ p\) norm for some approximation operators
• Probability error bounds for approximation of functions in reproducing kernel Hilbert spaces
• Kernel approximation on manifolds. II: The \(L_{\infty}\) norm of the \(L_2\) projector
• Sampled forms of functional PCA in reproducing kernel Hilbert spaces

This page was built for publication: Approximation properties of certain operator-induced norms on Hilbert spaces

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q765689)