Search

The paper considers suprema of empirical processes for linear time series indexed by functional classes. We derive an upper bound for the tail probability of the suprema under conditions on the size of the function class, the sample size, temporal dependence and the moment conditions of the underlying time series. Due to the dependence and heavy-tailness, our tail probability bound is substantially different from those classical exponential bounds obtained under the independence assumption in that it involves an extra polynomial decaying term. We allow both short- and long-range dependent processes. For empirical processes indexed by half intervals, our tail probability inequality is sharp up to a multiplicative constant. [ABSTRACT FROM AUTHOR]

We consider large scale multiple testing for data that have locally clustered signals. With this structure, we apply techniques from changepoint analysis and propose a boundary detection algorithm so that the clustering information can be utilized. Consequently the precision of the multiple testing procedure is substantially improved. We study tests with independent as well as dependent p-values. Monte Carlo simulations suggest that the methods perform well with realistic sample sizes and show improved detection ability compared with competing methods. Our procedure is applied to a genome-wide association dataset of blood lipids. [ABSTRACT FROM AUTHOR]

This paper considers the recursive estimation of quantiles using the stochastic gradient descent (SGD) algorithm with Polyak-Ruppert averaging. The algorithm offers a computationally and memory efficient alternative to the usual empirical estimator. Our focus is on studying the non-asymptotic behavior by providing exponentially decreasing tail probability bounds under mild assumptions on the smoothness of the density functions. This novel non-asymptotic result is based on a bound of the moment generating function of the SGD estimate. We apply our result to the problem of best arm identification in a multi-armed stochastic bandit setting under quantile preferences. [ABSTRACT FROM AUTHOR]

Based on the nonlinear system theory, we introduce previously undescribed dependence measures for stationary causal processes. Our physical and predictive dependence measures quantify the degree of dependence of outputs on inputs in physical systems. The proposed dependence measures provide a natural framework for a limit theory for stationary processes. In particular, under conditions with quite simple forms, we present limit theorems for partial sums, empirical processes, and kernel density estimates. The conditions are mild and easily verifiable because they are directly related to the data-generating mechanisms. [ABSTRACT FROM AUTHOR]

To study the roles of translational accuracy, translational efficiency, and the Hill-Robertson effect in codon usage bias, we studied the intragenic spatial distribution of synonymous codon usage bias in four prokaryotic (Escherichia coli, Bacillus subtilis, Sulfolobus tohodaii, and Thermotoga maritima) and two eukaryotic (Saccharomyces cerevisiae and Drosophila melanogaster) genomes. We generated supersequences at each codon position across genes in a genome and computed the overall bias at each codon position. By quantitatively evaluating the trend of spatial patterns using isotonic regression, we show that in yeast and prokaryotic genomes, codon usage bias increases along translational direction, which is consistent with purifying selection against nonsense errors. Fruit fly genes show a nearly symmetric M-shaped spatial pattern of codon usage bias, with less bias in the middle and both ends. The low codon usage bias in the middle region is best explained by interference (the Hill-Robertson effect) between selections at different codon positions. In both yeast and fruit fly, spatial patterns of codon usage bias are characteristically different from patterns of GC-content variations. Effect of expression level on the strength of codon usage bias is more conspicuous than its effect on the shape of the spatial distribution. [ABSTRACT FROM AUTHOR]

Estimation of an unstructured covariance matrix is difficult because of its positive-definiteness constraint. This obstacle is removed by regressing each variable on its predecessors, so that estimation of a covariance matrix is shown to be equivalent to that of estimating a sequence of varying-coefficient and varying-order regression models. Our framework is similar to the use of increasing-order autoregressive models in approximating the covariance matrix or the spectrum of a stationary time series. As an illustration, we adopt Fan & Zhang's (2000) two-step estimation of functional linear models and propose nonparametric estimators of covariance matrices which are guaranteed to be positive definite. For parsimony a suitable order for the sequence of (auto)regression models is found using penalised likelihood criteria like AIC and BIC. Some asymptotic results for the local polynomial estimators of components of a covariance matrix are established. Two longitudinal datasets are analysed to illustrate the methodology. A simulation study reveals the advantage of the nonparametric covariance estimator over the sample covariance matrix for large covariance matrices. [ABSTRACT FROM AUTHOR]

In this paper, we study the concentration property of stochastic gradient descent (SGD) solutions. In existing concentration analyses, researchers impose restrictive requirements on the gradient noise, such as boundedness or sub-Gaussianity. We consider a much richer class of noise where only finitely-many moments are required, thus allowing heavy-tailed noises. In particular, we obtain Nagaev type high-probability upper bounds for the estimation errors of averaged stochastic gradient descent (ASGD) in a linear model. Specifically, we prove that, after T steps of SGD, the ASGD estimate achieves an O(log(1=δ)/T + (δTq-1)-1/q) error rate with probability at least 1-δ, where q > 2 controls the tail of the gradient noise. In comparison, one has the O(log(1/δ)/T) error rate for sub-Gaussian noises. We also show that the Nagaev type upper bound is almost tight through an example, where the exact asymptotic form of the tail probability can be derived. Our concentration analysis indicates that, in the case of heavy-tailed noises, the polynomial dependence on the failure probability δ is generally unavoidable for the error rate of SGD. [ABSTRACT FROM AUTHOR]

We consider the problem of predicting aggregates or sums of future values of a process based on its past values. In contrast with the conventional prediction problem in which one predicts a future value given past values of the process, in our setting the number of aggregates can go to infinity with respect to the number of available observations. Consistency and Bahadur representations of the prediction estimators are established. A simulation study is carried out to assess the performance of different prediction estimators. [ABSTRACT FROM AUTHOR]

We consider nonparametric prediction problem for both short- and long-range-dependent linear processes. Asymptotic properties of local linear estimates are obtained and, for long-range-dependent processes, an interesting dichotomous phenomenon is described: the limiting distribution depends on the interplay between the strength of the dependence and the magnitude of the bandwidth. A simulation study is carried out to assess the performance of the nonparametric prediction estimator. [ABSTRACT FROM AUTHOR]

There is an increasing interest in studying time-varying quantiles, particularly for environmental processes. For instance, high pollution levels may cause severe respiratory problems, and large precipitation amounts can damage the environment, and have negative impacts on the society. In this article we address the problem of quantile curve estimation for a wide class of nonstationary and/or non-Gaussian processes. We discuss several nonparametric quantile curve estimates, give asymptotic results, and propose a data-driven procedure for the selection of smoothing parameters. This methodology provides a statistically reliable and computationally efficient graphical tool that can be used for the exploration and visualization of the behavior of time-varying quantiles for nonstationary time series. A Monte Carlo simulation study and two applications to ozone time series illustrate our method. R codes with the algorithm for selection of smoothing parameters (described in Section 3) are available in the online supplements. [ABSTRACT FROM AUTHOR]