Searcharxiv⌕ Search

arXiv subjects

Larry Goldstein

Publications and source records attributed to Larry Goldstein.

At least 37 records · Page 2Linked to original sources
arXiv2016

Structured signal recovery from non-linear and heavy-tailed measurements

We study high-dimensional signal recovery from non-linear measurements with design vectors having elliptically symmetric distribution. Special attention is devoted to the situation when the unknown signal belongs to a set of low statistical complexity, while both the measurements and the design vectors are heavy-tailed. We propose and analyze a new estimator that adapts to the structure of the problem, while being robust both to the possible model misspecification characterized by arbitrary non-linearity of the measurements as well as to data corruption modeled by the heavy-tailed distributions. Moreover, this estimator has low computational complexity. Our results are expressed in the form of exponential concentration inequalities for the error of the proposed estimator. On the technical side, our proofs rely on the generic chaining methods, and illustrate the power of this approach for statistical applications. Theory is supported by numerical experiments demonstrating that our estimator outperforms existing alternatives when data is heavy-tailed.

math.ST↗
arXiv2015

Functional van den Berg-Kesten-Reimer Inequalities and their Duals, with Applications

The BKR inequality conjectured by van den Berg and Kesten in [11], and proved by Reimer in [8], states that for $A$ and $B$ events on $S$, a finite product of finite sets $S_i,i=1,\ldots,n$, and $P$ any product measure on $S$, $$ P(A \Box B) \le P(A)P(B),$$ where the set $A \Box B$ consists of the elementary events which lie in both $A$ and $B$ for `disjoint reasons.' Precisely, with ${\bf n}:=\{1,\ldots,n\}$ and $K \subset {\bf n}$, for ${\bf x} \in S$ letting $[{\bf x}]_K=\{{\bf y} \in S: y_i = x_i, i \in K\}$, the set $A \Box B$ consists of all ${\bf x} \in S$ for which there exist disjoint subsets $K$ and $L$ of ${\bf n}$ for which $[{\bf x}]_K \subset A$ and $[{\bf x}]_L \subset B$. The BKR inequality is extended to the following functional version on a general finite product measure space $(S,\mathbb{S})$ with product probability measure $P$, $$E\left\{ \max_{\stackrel{K \cap L = \emptyset}{K \subset {\bf n}, L \subset {\bf n}}} \underline{f}_K({\bf X})\underline{g}_L({\bf X})\right\} \leq E\left\{f({\bf X})\right\}\,E\left\{g({\bf X})\right\},$$ where $f$ and $g$ are non-negative measurable functions, $\underline{f}_K({\bf x}) = {\rm ess} \inf_{{\bf y} \in [{\bf x}]_K}f({\bf y})$ and $\underline{g}_L({\bf x}) = {\rm ess} \inf_{{\bf y} \in [{\bf x}]_L}g({\bf y}).$ The original BKR inequality is recovered by taking $f({\bf x})={\bf 1}_A({\bf x})$ and $g({\bf x})={\bf 1}_B({\bf x})$, and applying the fact that in general ${\bf 1}_{A \Box B} \le \max_{K \cap L = \emptyset} \underline{f}_K({\bf x}) \underline{g}_L({\bf x})$. Related formulations, and functional versions of the dual inequality on events by Kahn, Saks, and Smyth [6], are also considered. Applications include order statistics, assignment problems, and paths in random graphs.

math.PR↗
arXiv2015

Stein's method and the rank distribution of random matrices over finite fields

With ${\mathcal{Q}}_{q,n}$ the distribution of $n$ minus the rank of a matrix chosen uniformly from the collection of all $n\times(n+m)$ matrices over the finite field $\mathbb{F}_q$ of size $q\ge2$, and ${\mathcal{Q}}_q$ the distributional limit of ${\mathcal{Q}}_{q,n}$ as $n\rightarrow\infty$, we apply Stein's method to prove the total variation bound $\frac{1}{8q^{n+m+1}}\leq\|{\mathcal{Q}}_{q,n}-{\mathcal{Q}}_q\|_{\mathrm{TV}}\leq\frac{3}{q^{n+m+1}}$. In addition, we obtain similar sharp results for the rank distributions of symmetric, symmetric with zero diagonal, skew symmetric, skew centrosymmetric and Hermitian matrices.

math.PR↗
arXiv2014

Gaussian Phase Transitions and Conic Intrinsic Volumes: Steining the Steiner Formula

Intrinsic volumes of convex sets are natural geometric quantities that also play important roles in applications, such as linear inverse problems with convex constraints, and constrained statistical inference. It is a well-known fact that, given a closed convex cone $C\subset \mathbb{R}^d$, its conic intrinsic volumes determine a probability measure on the finite set $\{0,1,...d\}$, customarily denoted by $\mathcal{L}(V_C)$. The aim of the present paper is to provide a Berry-Esseen bound for the normal approximation of ${\cal L}(V_C)$, implying a general quantitative central limit theorem (CLT) for sequences of (correctly normalised) discrete probability measures of the type $\mathcal{L}(V_{C_n})$, $n\geq 1$. This bound shows that, in the high-dimensional limit, most conic intrinsic volumes encountered in applications can be approximated by a suitable Gaussian distribution. Our approach is based on a variety of techniques, namely: (1) Steiner formulae for closed convex cones, (2) Stein's method and second order Poincaré inequality, (3) concentration estimates, and (4) Fourier analysis. Our results explicitly connect the sharp phase transitions, observed in many regularised linear inverse problems with convex constraints, with the asymptotic Gaussian fluctuations of the intrinsic volumes of the associated descent cones. In particular, our findings complete and further illuminate the recent breakthrough discoveries by Amelunxen, Lotz, McCoy and Tropp (2014) and McCoy and Tropp (2014) about the concentration of conic intrinsic volumes and its connection with threshold phenomena. As an additional outgrowth of our work we develop total variation bounds for normal approximations of the lengths of projections of Gaussian vectors on closed convex sets.

math.PR↗
arXiv2014

Concentration inequalities via zero bias couplings

The tails of the distribution of a mean zero, variance $σ^2$ random variable $Y$ satisfy concentration of measure inequalities of the form $\mathbb{P}(Y \ge t) \le \exp(-B(t))$ for $$ B(t)=\frac{t^2}{2( σ^2 + ct)} \quad \mbox{for $t \ge 0$, and} \quad B(t)=\frac{t}{c}\left( \log t - \log \log t - \frac{σ^2}{c}\right) \quad \mbox{for $t>e$} $$ whenever there exists a zero biased coupling of $Y$ bounded by $c$, under suitable conditions on the existence of the moment generating function of $Y$. These inequalities apply in cases where $Y$ is not a function of independent variables, such as for the Hoeffding statistic $Y=\sum_{i=1}^n a_{iπ(i)}$ where $A=(a_{ij})_{1 \le i,j \le n} \in \mathbb{R}^{n \times n}$ and the permutation $π$ has the uniform distribution over the symmetric group, and when its distribution is constant on cycle type.

math.PR↗
arXiv2013

Stein's method for the Beta distribution and the Pólya-Eggenberger Urn

Using a characterizing equation for the Beta distribution, Stein's method is applied to obtain bounds of the optimal order for the Wasserstein distance between the distribution of the scaled number of white balls drawn from a Pólya-Eggenberger urn and its limiting Beta distribution. The bound is computed by making a direct comparison between characterizing operators of the target and the Beta distribution, the former derived by extending Stein's density approach to discrete distributions. In addition, refinements are given to Döbler's result [12] for the Arcsine approximation for the fraction of time a simple random walk of even length spends positive, and so also to the distributions of its last return time to zero and its first visit to its terminal point, by supplying explicit constants to the present Wasserstein bound and also demonstrating that its rate is of the optimal order.

math.PR↗
arXiv2011

Clubbed Binomial Approximation for the Lightbulb Process

In the so called lightbulb process, on days r=1,..,n, out of n lightbulbs, all initially off, exactly r bulbs selected uniformly and independent of the past have their status changed from off to on, or vice versa. With W_n the number of bulbs on at the terminal time n and C_n a suitable clubbed binomial distribution, d_{TV}(W_n,C_n) \le 2.7314 \sqrt{n} e^{-(n+1)/3} for all n \ge 1. The result is shown using Stein's method.

math.PR↗
arXiv2011

Concentration of measures via size biased couplings

Let $Y$ be a nonnegative random variable with mean $μ$ and finite positive variance $σ^2$, and let $Y^s$, defined on the same space as $Y$, have the $Y$ size biased distribution, that is, the distribution characterized by E[Yf(Y)]=μE f(Y^s) for all functions $f$ for which these expectations exist. Under a variety of conditions on the coupling of Y and $Y^s$, including combinations of boundedness and monotonicity, concentration of measure inequalities hold. Examples include the number of relatively ordered subsequences of a random permutation, sliding window statistics including the number of m-runs in a sequence of coin tosses, the number of local maximum of a random function on a lattice, the number of urns containing exactly one ball in an urn allocation model, the volume covered by the union of $n$ balls placed uniformly over a volume n subset of d dimensional Euclidean space, the number of bulbs switched on at the terminal time in the so called lightbulb process, and the infinitely divisible and compound Poisson distributions that satisfy a bounded moment generating function condition.

math.PR↗
arXiv2011

Concentration of measure for the number of isolated vertices in the Erdős-Rényi random graph by size bias couplings

A concentration of measure result is proved for the number of isolated vertices $Y$ in the Erdős-Rényi random graph model on $n$ edges with edge probability $p$. When $μ$ and $σ^2$ denote the mean and variance of $Y$ respectively, $P((Y-μ)/σ\ge t)$ admits a bound of the form $e^{-kt^2}$ for some constant positive $k$ under the assumption $p \in (0,1)$ and $np\rightarrow c \in (0,\infty)$ as $n \rightarrow \infty$. The left tail inequality $$ P(\frac{Y-μ}σ\le -t)&\le& \exp(-\frac{t^2σ^2}{4μ}) $$ holds for all $n \in {2,3,...},p \in (0,1)$ and $t \ge 0$. The results are shown by coupling $Y$ to a random variable $Y^s$ having the $Y$-size biased distribution, that is, the distribution characterized by $E[Yf(Y)]=μE[f(Y^s)] $ for all functions $f$ for which these expectations exist.

math.PR↗
arXiv2011

Zero biasing and growth processes

The tools of zero biasing are adapted to yield a general result suitable for analyzing the behavior of certain growth processes. The main theorem is applied to prove central limit theorems, with explicit error terms in the L^1 metric, for certain statistics of the Jack measure on partitions and for the number of balls drawn in a Polya-Eggenberger urn process.

math.CO↗
arXiv2011

Stochastic comparisons of stratified sampling techniques for some Monte Carlo estimators

We compare estimators of the (essential) supremum and the integral of a function $f$ defined on a measurable space when $f$ may be observed at a sample of points in its domain, possibly with error. The estimators compared vary in their levels of stratification of the domain, with the result that more refined stratification is better with respect to different criteria. The emphasis is on criteria related to stochastic orders. For example, rather than compare estimators of the integral of $f$ by their variances (for unbiased estimators), or mean square error, we attempt the stronger comparison of convex order when possible. For the supremum, the criterion is based on the stochastic order of estimators.

math.ST↗
arXiv2011

On Optimal Allocation of a Continuous Resource Using an Iterative Approach and Total Positivity

We study a class of optimal allocation problems, including the well-known Bomber Problem, with the following common probabilistic structure. An aircraft equipped with an amount~$x$ of ammunition is intercepted by enemy airplanes arriving according to a homogenous Poisson process over a fixed time duration~$t$. Upon encountering an enemy, the aircraft has the choice of spending any amount~$0\le y\le x$ of its ammunition, resulting in the aircraft's survival with probability equal to some known increasing function of $y$. Two different goals have been considered in the literature concerning the optimal amount~$K(x,t)$ of ammunition spent: (i)~Maximizing the probability of surviving for time~$t$, which is the so-called Bomber Problem, and (ii) maximizing the number of enemy airplanes shot down during time~$t$, which we call the Fighter Problem. Several authors have attempted to settle the following conjectures about the monotonicity of $K(x,t)$: [A] $K(x,t)$ is decreasing in $t$, [B] $K(x,t)$ is increasing in $x$, and [C] the amount~$x-K(x,t)$ held back is increasing in $x$. [A] and [C] have been shown for the Bomber Problem with discrete ammunition, while [B] is still an open question. In this paper we consider both time and ammunition continuous, and for the Bomber Problem prove [A] and [C], while for the Fighter we prove [A] and [C] for one special case and [B] and [C] for another. These proofs involve showing that the optimal survival probability and optimal number shot down are totally positive of order 2 ($\mbox{TP}_2$) in the Bomber and Fighter Problems, respectively. The $\mbox{TP}_2$ property is shown by constructing convergent sequences of approximating functions through an iterative operation which preserves $\mbox{TP}_2$ and other properties.

math.PR↗
arXiv2011

A Berry Esseen Theorem for the Lightbulb Process

In the so called lightbulb process, on days $r=1,..., n$, out of $n$ lightbulbs, all initially off, exactly $r$ bulbs, selected uniformly and independent of the past, have their status changed from off to on, or vice versa. With $X$ the number of bulbs on at the terminal time $n$, an even integer, and $μ=n/2, σ^2=Var(X)$, we have $$ \sup_{z \in \mathbb{R}} |P(\frac{X-μ}σ \le z)-P(Z \le z)| \le \frac{n}{2σ^2} \barΔ_0 + 1.64 \frac{n}{σ^3}+ \frac{2}σ $$ where $Z$ is a standard normal random variable, and $$ \barΔ_0 = 1/2\sqrt{n}} + \frac{1}{2n} + 1/3 e^{-n/2} \qmq {for $n \ge 6$,} $$ yielding a bound of order $O(n^{-1/2})$ as $n \to \infty$. A similar, though slightly larger bound holds for $n$ odd. The results are shown using a version of Stein's method for bounded, monotone size bias couplings. The argument for even $n$ depends on the construction of a variable $X^s$ on the same space as $X$ that has the $X$-size bias distribution, that is, that satisfies \beas E [X g(X)] =μE[g(X^s)] \quad for all bounded continuous $g$, \enas and for which there exists a $B \ge 0$, in this case B=2, such that $X \le X^s \le X+B$ almost surely. The argument for $n$ odd is similar to that for $n$ even, but one first couples $X$ closely to $V$, a symmetrized version of $X$, for which a size bias coupling of $V$ to $V^s$ can proceed as in the even case. In both the even and odd cases, the crucial calculation of the variance of a conditional expectation requires detailed information on the spectral decomposition of the lightbulb chain.

math.PR↗
arXiv2010

Bounds on the constant in the mean central limit theorem

Let $X_1,\...,X_n$ be independent with zero means, finite variances $σ_1^2,\...,σ_n^2$ and finite absolute third moments. Let $F_n$ be the distribution function of $(X_1+\...+X_n)/σ$, where $σ^2=\sum_{i=1}^nσ_i^2$, and $Φ$ that of the standard normal. The $L^1$-distance between $F_n$ and $Φ$ then satisfies \[\Vert F_n-Φ\Vert_1\le\frac{1}{σ^3}\sum_{i=1}^nE|X_i|^3.\] In particular, when $X_1,\...,X_n$ are identically distributed with variance $σ^2$, we have \[\Vert F_n-Φ\Vert_1\le\frac{E|X_1|^3}{σ^3\sqrt{n}}\qquad for all $n\in\mathbb{N}$,\] corresponding to an $L^1$-Berry--Esseen constant of 1.

math.PR↗
arXiv2010

Normal approximation for coverage models over binomial point processes

We give error bounds which demonstrate optimal rates of convergence in the CLT for the total covered volume and the number of isolated shapes, for germ-grain models with fixed grain radius over a binomial point process of $n$ points in a toroidal spatial region of volume $n$. The proof is based on Stein's method via size-biased couplings.

math.PR↗
arXiv2010

Size bias, sampling, the waiting time paradox, and infinite divisibility: when is the increment independent?

With $X^*$ denoting a random variable with the $X$-size bias distribution, what are all distributions for $X$ such that it is possible to have $X^*=X+Y$, $Y\geq 0$, with $X$ and $Y$ {\em independent}? We give the answer, due to Steutel \cite{steutel}, and also discuss the relations of size biasing to the waiting time paradox, renewal theory, sampling, tightness and uniform integrability, compound Poisson distributions, infinite divisibility, and the lognormal distributions.

math.PR↗