A Graphical Goodness-of-Fit Test for Dependence Models in Higher Dimensions

This article introduces a graphical goodness-of-fit test for copulas in more than two dimensions. The test is based on pairs of variables and can thus be interpreted as a first-order approximation of the underlying dependence structure. The idea is to first transform pairs of data columns with the Rosenblatt transform to bivariate standard uniform distributions under the null hypothesis. This hypothesis can be graphically tested with a matrix of bivariate scatterplots, Q-Q plots, or other transformations. Furthermore, additional information can be encoded as background color, such as measures of association or (approximate) p-values of tests of independence. The proposed goodness-of-fit test is designed as a basic graphical tool for detecting deviations from a postulated, possibly high-dimensional, dependence model. Various examples are given and the methodology is applied to a financial dataset. An implementation is provided by the R package copula. Supplementary material for this article is available online, which provides the R package copula and reproduces all the graphical results of this article.     

Variable Selection Diagnostics Measures for High-Dimensional Regression

Markov chain Monte Carlo (MCMC) is nowadays a standard approach to numerical computation of integrals of the posterior density π of the parameter vector η. Unfortunately, Bayesian inference using MCMC is computationally intractable when the posterior density π is expensive to evaluate. In many such problems, it is possible to identify a minimal subvector β of η responsible for the expensive computation in the evaluation of π. We propose two approaches, DOSKA and INDA, that approximate π by interpolation in ways that exploit this computational structure to mitigate the curse of dimensionality. DOSKA interpolates π directly while INDA interpolates π indirectly by interpolating functions, for example, a regression function, upon which π depends. Our primary contribution is derivation of a Gaussian processes interpolant that provably improves over some of the existing approaches by reducing the effective dimension of the interpolation problem from dim(η) to dim(β). This allows a dramatic reduction of the number of expensive evaluations necessary to construct an accurate approximation of π when dim(η) is high but dim(β) is low.

We illustrate the proposed approaches in a case study for a spatio-temporal linear model for air pollution data in the greater Boston area.

Supplemental materials include proofs, details, and software implementation of the proposed procedures.     

Combining an Additive and Tree-Based Regression Model Simultaneously: STIMA

Additive models and tree-based regression models are two main classes of statistical models used to predict the scores on a continuous response variable. It is known that additive models become very complex in the presence of higher order interaction effects, whereas some tree-based models, such as CART, have problems capturing linear main effects of continuous predictors. To overcome these drawbacks, the regression trunk model has been proposed: a multiple regression model with main effects and a parsimonious amount of higher order interaction effects. The interaction effects can be represented by a small tree: a regression trunk. This article proposes a new algorithm—Simultaneous Threshold Interaction Modeling Algorithm (STIMA)—to estimate a regression trunk model that is more general and more efficient than the initial one (RTA) and is implemented in the R-package stima. Results from a simulation study show that the performance of STIMA is satisfactory for sample sizes of 200 or higher. For sample sizes of 300 or higher, the 0.50 SE rule is the best pruning rule for a regression trunk in terms of power and Type I error. For sample sizes of 200, the 0.80 SE rule is recommended. Results from a comparative study of eight regression methods applied to ten benchmark datasets suggest that STIMA and GUIDE are the best performers in terms of cross-validated prediction error. STIMA appeared to be the best method for datasets containing many categorical variables. The characteristics of a regression trunk model are illustrated using the Boston house price dataset.

Supplemental materials for this article, including the R-package stima, are available online.     

Local Derivative-Free Approximation of Computationally Expensive Posterior Densities

Bayesian inference using Markov chain Monte Carlo (MCMC) is computationally prohibitive when the posterior density of interest, π, is computationally expensive to evaluate. We develop a derivative-free algorithm GRIMA to accurately approximate π by interpolation over its high-probability density (HPD) region, which is initially unknown. Our local approach reduces the waste of computational budget on approximation of π in the low-probability region, which is inherent in global experimental designs. However, estimation of the HPD region is nontrivial when derivatives of π are not available or are not informative about the shape of the HPD region. Without relying on derivatives, GRIMA iterates (a) sequential knot selection over the estimated HPD region of π to refine the surrogate posterior and (b) re-estimation of the HPD region using an MCMC sample from the updated surrogate density, which is inexpensive to obtain. GRIMA is applicable to approximation of general unnormalized posterior densities. To determine the range of tractable problem dimensions, we conduct simulation experiments on test densities with linear and nonlinear component-wise dependence, skewness, kurtosis and multimodality. Subsequently, we use GRIMA in a case study to calibrate a computationally intensive nonlinear regression model to real data from the Town Brook watershed. Supplemental materials for this article are available online.     

Convex Modeling of Interactions With Strong Heredity

We consider the task of fitting a regression model involving interactions among a potentially large set of covariates, in which we wish to enforce strong heredity. We propose FAMILY, a very general framework for this task. Our proposal is a generalization of several existing methods, such as VANISH, hierNet, the all-pairs lasso, and the lasso using only main effects. It can be formulated as the solution to a convex optimization problem, which we solve using an efficient alternating directions method of multipliers (ADMM) algorithm. This algorithm has guaranteed convergence to the global optimum, can be easily specialized to any convex penalty function of interest, and allows for a straightforward extension to the setting of generalized linear models. We derive an unbiased estimator of the degrees of freedom of FAMILY, and explore its performance in a simulation study and on an HIV sequence dataset. Supplementary materials for this article are available online.     

On Skew Power Serieswise Armendariz Rings

For a ring endomorphism α, we introduce and investigate SPA-rings which are a generalization of α-rigid rings and determine the radicals of the skew polynomial rings R[x; α], R[x, x ^?1; α] and the skew power series rings R[[x; α]], R[[x, x ^?1; α]], in terms of those of R. We prove that several properties transfer between R and the extensions, in case R is an SPA-ring. We will construct various types of nonreduced SPA-rings and show SPA is a strictly stronger condition than α-rigid.     

State Estimation Schemes for Independent Component Coupled Hidden Markov Models

Conventional Hidden Markov models generally consist of a Markov chain observed through a linear map corrupted by additive noise. This general class of model has enjoyed a huge and diverse range of applications, for example, speech processing, biomedical signal processing and more recently quantitative finance. However, a lesser known extension of this general class of model is the so-called Factorial Hidden Markov Model (FHMM). FHMMs also have diverse applications, notably in machine learning, artificial intelligence and speech recognition [13 Ghahramani , Z. , and Jordan , M. 1997 . Factorial hidden Markov models . Machine Learning 29 : 245 – 273 .[Crossref], [Web of Science ®] , [Google Scholar], 17 Logan , B. , and Moreno , P.J. 1997 . Factorial Hidden Markov Models for Speech Recognition: Preliminary Experiments. Cambridge Research Laboratory, Technical Report CRL 97/7, September.  [Google Scholar]]. FHMMs extend the usual class of HMMs, by supposing the partially observed state process is a finite collection of distinct Markov chains, either statistically independent or dependent. There is also considerable current activity in applying collections of partially observed Markov chains to complex action recognition problems, see, for example, [6 Brand , M. , Oliver , N. , and Pentland , A. 1997 . Coupled hidden Markov models for complex action recognition . IEEE Conference on Computer Vision and Pattern Recognition , San Juan , Puerto Rico . [Google Scholar]].

In this article we consider the Maximum Likelihood (ML) parameter estimation problem for FHMMs. Much of the extant literature concerning this problem presents parameter estimation schemes based on full data log-likelihood EM algorithms. This approach can be slow to converge and often imposes heavy demands on computer memory. The latter point is particularly relevant for the class of FHMMs where state space dimensions are relatively large.

The contribution in this article is to develop new recursive formulae for a filter-based EM algorithm that can be implemented online. Our new formulae are equivalent ML estimators, however, these formulae are purely recursive and so, significantly reduce numerical complexity and memory requirements. A computer simulation is included to demonstrate the performance of our results.     

Bounds on Lifting Continuous-State Markov Chains to Speed Up Mixing

It is often possible to speed up the mixing of a Markov chain \(\{ X_{t} \}_{t \in \mathbb {N}}\) on a state space \(\Omega \) by lifting, that is, running a more efficient Markov chain \(\{ \widehat{X}_{t} \}_{t \in \mathbb {N}}\) on a larger state space \(\hat{\Omega } \supset \Omega \) that projects to \(\{ X_{t} \}_{t \in \mathbb {N}}\) in a certain sense. Chen et al. (Proceedings of the 31st annual ACM symposium on theory of computing. ACM, 1999) prove that for Markov chains on finite state spaces, the mixing time of any lift of a Markov chain is at least the square root of the mixing time of the original chain, up to a factor that depends on the stationary measure of \(\{X_t\}_{t \in \mathbb {N}}\). Unfortunately, this extra factor makes the bound in Chen et al. (1999) very loose for Markov chains on large state spaces and useless for Markov chains on continuous state spaces. In this paper, we develop an extension of the evolving set method that allows us to refine this extra factor and find bounds for Markov chains on continuous state spaces that are analogous to the bounds in Chen et al. (1999). These bounds also allow us to improve on the bounds in Chen et al. (1999) for some chains on finite state spaces.     

Automorphism Group of q-Quantum Torus Lie Algebra with q a Root of Unity*

Let α be an automorphism of a ring R. We study the skew Armendariz of Laurent series type rings (α-LA rings), as a generalization of the standard Armendariz condition from polynomials to skew Laurent series. We study on the relationship between the Baerness and p.p. property of a ring R and these of the skew Laurent series ring R[[x, x ^?1; α]], in case R is an α-LA ring. Moreover, we prove that for an α-weakly rigid ring R, R[[x, x ^?1; α]] is a left p.q.-Baer ring if and only if R is left p.q.-Baer and every countable subset of S _?(R) has a generalized countable join in R. Various types of examples of α-LA rings are provided.     

LU-Factorization Versus Wiener-Hopf Factorization for Markov Chains

Our initial motivation was to understand links between Wiener-Hopf factorizations for random walks and LU-factorizations for Markov chains as interpreted by Grassman (Eur. J. Oper. Res. 31(1):132–139, 1987). Actually, the first ones are particular cases of the second ones, up to Fourier transforms. To show this, we produce a new proof of LU-factorizations which is valid for any Markov chain with a denumerable state space equipped with a pre-order relation. Factors have nice interpretations in terms of subordinated Markov chains. In particular, the LU-factorization of the potential matrix determines the law of the global minimum of the Markov chain. For any matrix, there are two main LU-factorizations according as you decide to enter 1 in the diagonal of the first or of the second factor. When we factorize the generator of a Markov chain, one factorization is always valid while the other requires some hypothesis on the graph of the transition matrix. This dissymmetry comes from the fact that the class of sub-stochastic matrices is not stable under transposition. We generalize our work to the class of matrices with spectral radius less than one; this allows us to play with transposition and thus with time-reversal. We study some particular cases such as: skip-free Markov chains, random walks (this gives the WH-factorization), reversible Markov chains (this gives the Cholesky factorization). We use the LU-factorization to compute invariant measures. We present some pathologies: non-associativity, non-unicity; these can be cured by smooth assumptions (e.g. irreductibility).     

The Block Criterion for Multiscale Inference About a Density,With Applications to Other Multiscale Problems

The use of multiscale statistics, that is, the simultaneous inference about various stretches of data via multiple localized statistics, is a natural and popular method for inference about, for example, local qualitative characteristics of a regression function, a density, or its hazard rate. We focus on the problem of providing simultaneous confidence statements for the existence of local increases and decreases of a density and address several statistical and computational issues concerning such multiscale statistics. We first review the benefits of employing scale-dependent critical values for multiscale statistics and then derive an approximation scheme that results in a fast algorithm while preserving statistical optimality properties. The main contribution is a methodology for calibrating multiscale statistics that does not require a case-by-case derivation of its specific form. We show that in the above density context the methodology possesses statistical optimality properties and allows for a fast algorithm. We illustrate the methodology with two further examples: a multiscale statistic introduced by Gijbels and Heckman for inference about a hazard rate and local rank tests introduced by Dümbgen for inference in nonparametric regression.

Code for the density application is available as the R package modehunt on CRAN. Additional code to compute critical values, reproduce the hazard rate and local rank example and the plots in the paper as well as datasets containing simulation results and an appendix with all the proofs of the theorems are available online as supplemental material.     

Markov Limid processes for representing and solving renewal problems

In this paper a new tool for simultaneous optimisation of decisions on multiple time scales is presented. The tool combines the dynamic properties of Markov decision processes with the flexible and compact state space representation of LImited Memory Influence Diagrams (Limids). A temporal version of Limids, TemLimids, is defined by adding time-related functions to utility nodes. As a result, expected discounted utility, as well as expected relative utility might be used as optimisation criteria in TemLimids. Optimisation proceeds as in ordinary Limids. A sequence of such TemLimids can be used to model a Markov Limid Process, where each TemLimid represents a macro action. Algorithms are presented to find optimal plans for a sequence of such macro actions. Use of algorithms is illustrated based on an extended version of an example from pig production originally used to introduce the Limid concept.     

Fluctuation Theory for Markov Random Walks

Two fundamental theorems by Spitzer–Erickson and Kesten–Maller on the fluctuation-type (positive divergence, negative divergence or oscillation) of a real-valued random walk \((S_{n})_{n\ge 0}\) with iid increments \(X_{1},X_{2},\ldots \) and the existence of moments of various related quantities like the first passage into \((x,\infty )\) and the last exit time from \((-\infty ,x]\) for arbitrary \(x\ge 0\) are studied in the Markov-modulated situation when the \(X_{n}\) are governed by a positive recurrent Markov chain \(M=(M_{n})_{n\ge 0}\) on a countable state space \(\mathcal {S}\); thus, for a Markov random walk \((M_{n},S_{n})_{n\ge 0}\). Our approach is based on the natural strategy to draw on the results in the iid case for the embedded ordinary random walks \((S_{\tau _{n}(i)})_{n\ge 0}\), where \(\tau _{1}(i),\tau _{2}(i),\ldots \) denote the successive return times of M to state i, and an analysis of the excursions of the walk between these epochs. However, due to these excursions, generalizations of the aforementioned theorems are surprisingly more complicated and require the introduction of various excursion measures so as to characterize the existence of moments of different quantities.     

Skorokhod embeddings for two-sided Markov chains

Let \((X_n :n\in \mathbb {Z})\) be a two-sided recurrent Markov chain with fixed initial state \(X_0\) and let \(\nu \) be a probability measure on its state space. We give a necessary and sufficient criterion for the existence of a non-randomized time T such that \((X_{T+n} :n\in \mathbb {Z})\) has the law of the same Markov chain with initial distribution \(\nu \). In the case when our criterion is satisfied we give an explicit solution, which is also a stopping time, and study its moment properties. We show that this solution minimizes the expectation of \(\psi (T)\) in the class of all non-negative solutions, simultaneously for all non-negative concave functions \(\psi \).     

Numerical approximations for nonlinear stochastic systems with delays

We extend the numerical methods of [Kushner, H.J. and Dupuis, P., 1992 Kushner, H.J. and Dupuis, P. 2001. Numerical Methods for Stochastic Control Problems in Continuous Time, 2nd ed., Berlin and New York: Springer-Verlag. [Crossref] , [Google Scholar], Numerical Methods for Stochastic Control Problems in Continuous Time, 2nd ed., 2001 (Berlin and New York: Springer Verlag], known as the Markov chain approximation methods, to controlled general nonlinear delayed reflected diffusion models. Both the path and the control can be delayed. For the no-delay case, the method covers virtually all models of current interest. The method is robust, the approximations have physical interpretations as control problems closely related to the original one, and there are many effective methods for getting the approximations, and for solving the Bellman equation for low-dimensional problems. These advantages carry over to the delay problem. It is shown how to adapt the methods for getting the approximations, and the convergence proofs are outlined for the discounted cost function. Extensions to all of the cost functions of current interest as well as to models with Poisson jump terms are possible. The paper is particularly concerned with representations of the state and algorithms that minimize the memory requirements.     

Independent sampling for Bayesian normal conditional autoregressive models with OpenCL acceleration

A new computational strategy produces independent samples from the joint posterior distribution for a broad class of Bayesian spatial and spatiotemporal conditional autoregressive models. The method is based on reparameterization and marginalization of the posterior distribution and massive parallelization of rejection sampling using graphical processing units (GPUs) or other accelerators. It enables very fast sampling for small to moderate-sized datasets (up to approximately 10,000 observations) and feasible sampling for much larger datasets. Even using a mid-range GPU and a high-end CPU, the GPU-based implementation is up to 30 times faster than the same algorithm run serially on a single CPU, and the numbers of effective samples per second are orders of magnitude higher than those obtained with popular Markov chain Monte Carlo software. The method has been implemented in the R package CARrampsOcl. This work provides both a practical computing strategy for fitting a popular class of Bayesian models and a proof of concept that GPU acceleration can make independent sampling from Bayesian joint posterior densities feasible.     

Modelling,simulation, animation,and real-time control (MoSART) for a class of electromechanical systems: a system-theoretic approach

This paper describes an interactive modelling, simulation, animation, and real-time control (MoSART) environment for a class of ‘cart-pendulum’ electromechanical systems that may be used to enhance learning within differential equations and linear algebra classes. The environment is useful for conveying fundamental mathematical/systems concepts through computer-aided analysis, design, graphical visualization, and 3D animation. Referred to as Cart-Pendulum Control3D-Lab1Control3D-Lab-short for control 3D animation laboratory. The author to whom correspondence should be addressed., the environment is based on Microsoft Windows/Visual C++/Direct-3D and MATLAB/Simulink2MATLAB and Simulant are trademarks of The Masterworks, Inc.. Pull-down menus provide access to systems/models/control laws, exogenous signals, parameters, animation models and visual indicators, a suite of (easy-to-modify) Simulant diagrams containing models and control laws, MATLAB m-files for detailed analysis and design, and detailed documentation for each of the above. Three blocks enable animation, joystick inputs, and real-time animation within Simulant. Examples are presented to illustrate the utility of the environment as a valuable tool for analysing/visualizing the above class of electromechanical systems and for enhancing mathematics instruction.     

Functional Central Limit Theorems for Supercritical Superprocesses

In this paper, we establish some functional central limit theorems for a large class of general supercritical superprocesses with spatially dependent branching mechanisms satisfying a second moment condition. In the particular case when the state \(E\) is a finite set and the underlying motion is an irreducible Markov chain on \(E\), our results are superprocess analogs of the functional central limit theorems of Janson (Stoch. Process. Appl. 110:177–245, 2004) for supercritical multitype branching processes. The results of this paper are refinements of the central limit theorems in Ren et al. (Stoch. Process. Appl. 125:428–457, 2015).     

Combinatorial Markov chains on linear extensions

We consider generalizations of Schützenberger’s promotion operator on the set $\mathcal{L}$ of linear extensions of a finite poset of size n. This gives rise to a strongly connected graph on $\mathcal{L}$ . By assigning weights to the edges of the graph in two different ways, we study two Markov chains, both of which are irreducible. The stationary state of one gives rise to the uniform distribution, whereas the weights of the stationary state of the other have a nice product formula. This generalizes results by Hendricks on the Tsetlin library, which corresponds to the case when the poset is the anti-chain and hence $\mathcal{L}=S_{n}$ is the full symmetric group. We also provide explicit eigenvalues of the transition matrix in general when the poset is a rooted forest. This is shown by proving that the associated monoid is $\mathcal {R}$ -trivial and then using Steinberg’s extension of Brown’s theory for Markov chains on left regular bands to $\mathcal {R}$ -trivial monoids.     

Convergence of symmetric Markov chains on {\mathbb{Z}^d}

For each n let ${Y^{(n)}_t}$ be a continuous time symmetric Markov chain with state space ${n^{-1} \mathbb{Z}^d}$ . Conditions in terms of the conductances are given for the convergence of the ${Y^{(n)}_t}$ to a symmetric Markov process Y _t on ${\mathbb{R}^d}$ . We have weak convergence of $\{{Y^{(n)}_t: t \leq t_0\}}$ for every t ₀ and every starting point. The limit process Y has a continuous part and may also have jumps.