arXiv subjects

Richard Nickl

Publications and source records attributed to Richard Nickl.

47 records · Page 3Linked to original sources

arXiv2012

Rates of contraction for posterior distributions in $\bolds{L^r}$-metrics, $\bolds{1\le r\le\infty}$

The frequentist behavior of nonparametric Bayes estimates, more specifically, rates of contraction of the posterior distributions to shrinking $L^r$-norm neighborhoods, $1\le r\le\infty$, of the unknown parameter, are studied. A theorem for nonparametric density estimation is proved under general approximation-theoretic assumptions on the prior. The result is applied to a variety of common examples, including Gaussian process, wavelet series, normal mixture and histogram priors. The rates of contraction are minimax-optimal for $1\le r\le2$, but deteriorate as $r$ increases beyond 2. In the case of Gaussian nonparametric regression a Gaussian prior is devised for which the posterior contracts at the optimal rate in all $L^r$-norms, $1\le r\le\infty$.

math.ST↗

arXiv2012

On adaptive inference and confidence bands

The problem of existence of adaptive confidence bands for an unknown density $f$ that belongs to a nested scale of Hölder classes over $\mathbb{R}$ or $[0,1]$ is considered. Whereas honest adaptive inference in this problem is impossible already for a pair of Hölder balls $Σ(r),Σ(s),r\ne s$, of fixed radius, a nonparametric distinguishability condition is introduced under which adaptive confidence bands can be shown to exist. It is further shown that this condition is necessary and sufficient for the existence of honest asymptotic confidence bands, and that it is strictly weaker than similar analytic conditions recently employed in Giné and Nickl [Ann. Statist. 38 (2010) 1122--1170]. The exceptional sets for which honest inference is not possible have vanishingly small probability under natural priors on Hölder balls $Σ(s)$. If no upper bound for the radius of the Hölder balls is known, a price for adaptation has to be paid, and near-optimal adaptation is possible for standard procedures. The implications of these findings for a general theory of adaptive inference are discussed.

math.ST↗

arXiv2012

Spatially Adaptive Density Estimation by Localised Haar Projections

Given a random sample from some unknown density $f_0: \mathbb R \to [0, \infty)$ we devise Haar wavelet estimators for $f_0$ with variable resolution levels constructed from localised test procedures (as in Lepski, Mammen, and Spokoiny (1997, Ann. Statist.)). We show that these estimators adapt to spatially heterogeneous smoothness of $f_0$, simultaneously for every point $x$ in a fixed interval, in sup-norm loss. The thresholding constants involved in the test procedures can be chosen in practice under the idealised assumption that the true density is locally constant in a neighborhood of the point $x$ of estimation, and an information theoretic justification of this practice is given.

math.ST↗

arXiv2011

Global uniform risk bounds for wavelet deconvolution estimators

We consider the statistical deconvolution problem where one observes $n$ replications from the model $Y=X+ε$, where $X$ is the unobserved random signal of interest and $ε$ is an independent random error with distribution $ϕ$. Under weak assumptions on the decay of the Fourier transform of $ϕ,$ we derive upper bounds for the finite-sample sup-norm risk of wavelet deconvolution density estimators $f_n$ for the density $f$ of $X$, where $f:\mathbb{R}\to \mathbb{R}$ is assumed to be bounded. We then derive lower bounds for the minimax sup-norm risk over Besov balls in this estimation problem and show that wavelet deconvolution density estimators attain these bounds. We further show that linear estimators adapt to the unknown smoothness of $f$ if the Fourier transform of $ϕ$ decays exponentially and that a corresponding result holds true for the hard thresholding wavelet estimator if $ϕ$ decays polynomially. We also analyze the case where $f$ is a "supersmooth"/analytic density. We finally show how our results and recent techniques from Rademacher processes can be applied to construct global confidence bands for the density $f$.

math.ST↗

arXiv2011

Concentration Inequalities and Confidence Bands for Needlet Density Estimators on Compact Homogeneous Manifolds

Let $X_1,...,X_n$ be a random sample from some unknown probability density $f$ defined on a compact homogeneous manifold $\mathbf M$ of dimension $d \ge 1$. Consider a 'needlet frame' $\{ϕ_{j η}\}$ describing a localised projection onto the space of eigenfunctions of the Laplace operator on $\mathbf M$ with corresponding eigenvalues less than $2^{2j}$, as constructed in \cite{GP10}. We prove non-asymptotic concentration inequalities for the uniform deviations of the linear needlet density estimator $f_n(j)$ obtained from an empirical estimate of the needlet projection $\sum_ηϕ_{j η} \int f ϕ_{j η}$ of $f$. We apply these results to construct risk-adaptive estimators and nonasymptotic confidence bands for the unknown density $f$. The confidence bands are adaptive over classes of differentiable and H\"{older}-continuous functions on $\mathbf M$ that attain their Hölder exponents.

math.ST↗

arXiv2011

Adaptive estimation of a distribution function and its density in sup-norm loss by wavelet and spline projections

Given an i.i.d. sample from a distribution $F$ on $\mathbb{R}$ with uniformly continuous density $p_0$, purely data-driven estimators are constructed that efficiently estimate $F$ in sup-norm loss and simultaneously estimate $p_0$ at the best possible rate of convergence over Hölder balls, also in sup-norm loss. The estimators are obtained by applying a model selection procedure close to Lepski's method with random thresholds to projections of the empirical measure onto spaces spanned by wavelets or $B$-splines. The random thresholds are based on suprema of Rademacher processes indexed by wavelet or spline projection kernels. This requires Bernstein-type analogs of the inequalities in Koltchinskii [Ann. Statist. 34 (2006) 2593-2656] for the deviation of suprema of empirical processes from their Rademacher symmetrizations.

math.ST↗

arXiv2010

Efficient Simulation-Based Minimum Distance Estimation and Indirect Inference

Given a random sample from a parametric model, we show how indirect inference estimators based on appropriate nonparametric density estimators (i.e., simulation-based minimum distance estimators) can be constructed that, under mild assumptions, are asymptotically normal with variance-covarince matrix equal to the Cramer-Rao bound.

math.ST↗

arXiv2010

Confidence bands in density estimation

Given a sample from some unknown continuous density $f:\mathbb{R}\to\mathbb{R}$, we construct adaptive confidence bands that are honest for all densities in a "generic" subset of the union of $t$-Hölder balls, $0<t\le r$, where $r$ is a fixed but arbitrary integer. The exceptional ("nongeneric") set of densities for which our results do not hold is shown to be nowhere dense in the relevant Hölder-norm topologies. In the course of the proofs we also obtain limit theorems for maxima of linear wavelet and kernel density estimators, which are of independent interest.

math.ST↗

arXiv2009

Uniform limit theorems for wavelet density estimators

Let $p_n(y)=\sum_k\hatα_kϕ(y-k)+\sum_{l=0}^{j_n-1}\sum_k\hat β_{lk}2^{l/2}ψ(2^ly-k)$ be the linear wavelet density estimator, where $ϕ$, $ψ$ are a father and a mother wavelet (with compact support), $\hatα_k$, $\hatβ_{lk}$ are the empirical wavelet coefficients based on an i.i.d. sample of random variables distributed according to a density $p_0$ on $\mathbb{R}$, and $j_n\in\mathbb{Z}$, $j_n\nearrow\infty$. Several uniform limit theorems are proved: First, the almost sure rate of convergence of $\sup_{y\in\mathbb{R}}|p_n(y)-Ep_n(y)|$ is obtained, and a law of the logarithm for a suitably scaled version of this quantity is established. This implies that $\sup_{y\in\mathbb{R}}|p_n(y)-p_0(y)|$ attains the optimal almost sure rate of convergence for estimating $p_0$, if $j_n$ is suitably chosen. Second, a uniform central limit theorem as well as strong invariance principles for the distribution function of $p_n$, that is, for the stochastic processes $\sqrt{n}(F_n ^W(s)-F(s))=\sqrt{n}\int_{-\infty}^s(p_n-p_0),s\in\mathbb{R}$, are proved; and more generally, uniform central limit theorems for the processes $\sqrt{n}\int(p_n-p_0)f$, $f\in\mathcal{F}$, for other Donsker classes $\mathcal{F}$ of interest are considered. As a statistical application, it is shown that essentially the same limit theorems can be obtained for the hard thresholding wavelet estimator introduced by Donoho et al. [Ann. Statist. 24 (1996) 508--539].

math.ST↗

arXiv2008

A simple adaptive estimator of the integrated square of a density

Given an i.i.d. sample $X_1,...,X_n$ with common bounded density $f_0$ belonging to a Sobolev space of order $α$ over the real line, estimation of the quadratic functional $\int_{\mathbb{R}}f_0^2(x) \mathrm{d}x$ is considered. It is shown that the simplest kernel-based plug-in estimator \[\frac{2}{n(n-1)h_n}\sum_{1\leq i 1/4$ and rate-optimal if $α\le1/4$. A data-driven rule to choose the bandwidth $h_n$ is then proposed, which does not depend on prior knowledge of $α$, so that the corresponding estimator is rate-adaptive for $α\leq1/4$ and asymptotically efficient if $α>1/4$.

math.ST↗

arXiv2006

Empirical and Gaussian processes on Besov classes

We give several conditions for pregaussianity of norm balls of Besov spaces defined over $\mathbb{R}^d$ by exploiting results in Haroske and Triebel (2005). Furthermore, complementing sufficient conditions in Nickl and Pötscher (2005), we give necessary conditions on the parameters of the Besov space to obtain the Donsker property of such balls. For certain parameter combinations Besov balls are shown to be pregaussian but not Donsker.

math.PR↗