Asymptotic properties of subspace estimators

Since the proposal of subspace methods in the 1980s and early 1990s substantial efforts have been made in the analysis of the statistical properties of the algorithms. This paper surveys the literature on the asymptotic properties of particular subspace methods used for linear, dynamic, time invariant, discrete time systems. The goals of this paper are threefold: First this survey tries to present the most relevant results on the asymptotic properties of estimators obtained using subspace methods. Secondly the main methods and tools that have been used in the derivation of these results are presented to make the literature more accessible. Thirdly main unsolved questions and rewarding research topics are identified some of which can be attacked using the toolbox discussed in the paper.     

Asymptotic properties of mixture-of-experts models

The statistical properties of the likelihood ratio test statistic (LRTS) for mixture-of-expert models are addressed in this paper. This question is essential when estimating the number of experts in the model. Our purpose is to extend the existing results for simple mixture models (Liu and Shao, 2003 [8]) and mixtures of multilayer perceptrons (Olteanu and Rynkiewicz, 2008 [9]). In this paper we first study a simple example which embodies all the difficulties arising in such models. We find that in the most general case the LRTS diverges but, with additional assumptions, the behavior of such models can be totally explicated.     

Asymptotic properties of the Fisher kernel

This letter analyzes the Fisher kernel from a statistical point of view. The Fisher kernel is a particularly interesting method for constructing a model of the posterior probability that makes intelligent use of unlabeled data (i.e., of the underlying data density). It is important to analyze and ultimately understand the statistical properties of the Fisher kernel. To this end, we first establish sufficient conditions that the constructed posterior model is realizable (i.e., it contains the true distribution). Realizability immediately leads to consistency results. Subsequently, we focus on an asymptotic analysis of the generalization error, which elucidates the learning curves of the Fisher kernel and how unlabeled data contribute to learning. We also point out that the squared or log loss is theoretically preferable-because both yield consistent estimators-to other losses such as the exponential loss, when a linear classifier is used together with the Fisher kernel. Therefore, this letter underlines that the Fisher kernel should be viewed not as a heuristics but as a powerful statistical tool with well-controlled statistical properties.     

Asymptotic properties of incrementally stable systems

It is shown that incremental stability of an input-output operator ensures asymptotic stability of any equilibrium pair (x_e, u _e) of its state representation if suitable minimality assumptions hold     

Asymptotic properties of discrete unitary transforms

A method for studying the asymptotic behavior of discrete transformations is developed using numerical quadrature theory. This method allows a more convenient examination of the correlation properties of common unitary transforms for large block sizes. As a practical result of this method it is shown that the discrete cosine transform is asymptotically optimal for all finite-order Markov signals.     

Asymptotic properties of set membership identification algorithms

This paper addresses the asymptotic worst-case properties of set membership identification (SMID) algorithms. We first present a set membership identification algorithm which can be used with a model structure consisting of parametric and nonparametric uncertainty, as well as output additive disturbances. This algorithm is then studied in the context of asymptotic worst-case behavior. We derive lower bounds on the worst-case achievable identification error measured by the volume, as well as the sum-of-sidelengths of the identified ellipsoidal uncertainty sets. We then show that there exist inputs which can shrink the uncertainty sets to these lower bounds.     

Asymptotic properties of stochastic timed event graphs

This paper addresses the asymptotic properties of stochastic timed event graphs. The transition firing times are random variables with general distribution. Asymptotic properties with respect to the net structure, the transition firing times and the initial marking are established on the basis of some new performance bounds. Applications to manufacturing systems are presented. We prove in particular a conjecture which claims that the throughput rate of a transfer line decreases to a positive value when the number of machines increases.     

Asymptotic properties of least-squares estimates of Hammerstein-Wiener models

This paper investigates the asymptotic properties of least squares estimates of Hammerstein-Wiener model structures, and in doing so establishes consistency and asymptotic normality under fairly mild conditions on the additive noise process, the inputs and the static non-linearities. In relation to the asymptotic distributional results, a consistent procedure for the estimation of the asymptotic variance of the parameter estimates is provided. A key theme of this paper is to demonstrate how recent results from the econometrics literature may be employed in an engineering setting. In this respect the Hammerstein-Wiener model structure serves as a demonstration example. A simulation study complements the theoretical findings.     

Asymptotic properties of black-box identification of transfer functions

The problem of estimating the transfer function of a linear, stochastic system is considered. The transfer function is parametrized as a black box and no given order is chosen a priori. This means that the model orders may increase to infinity when the number of observed data tends to infinity. The consistency and convergence properties of the resulting transfer function estimates are investigated. Asymptotic expressions for the variances and distributions of these estimates are also derived for the case that the model orders increase. It is shown that the variance of the transfer function estimate at a certain frequency is asymptotically given by the noise-to-signal ratio at that frequency mulliplied by the model-order-to-number-of-data-points ratio.     

Asymptotic properties of sign algorithms for adaptive filtering

This paper develops asymptotic properties of a class of sign-error algorithms with expanding truncation bounds for adaptive filtering. Under merely stationary ergodicity and finite second moments of the reference and output signals, and using trajectory-subsequence (TS) method, it is proved that the algorithm convergers almost surely. Then, a mean squares estimate is derived for the estimation error and a suitably scaled sequence of the estimation error is shown to converge to a diffusion process. The scaling factor together with the stationary covariance gives the rate of convergence result. Moreover, an algorithm under mean squares criterion with expanding truncation bounds is also examined. Compared with the existing results in the literature, sufficient conditions for almost sure convergence are much relaxed. A simple example is provided for demonstration purpose.     

Asymptotic properties of minimax trees and game-searching procedures

The model most frequently used for evaluating the behavior of game-searching methods consists of a uniform tree of height h and a branching degree d, where the terminal positions are assigned random, independent and identically distributed values. This paper highlights some curious properties of such trees when h is very large and examines their implications on the complexity of various game-searching methods.If the terminal positions are assigned a WIN-LOSS status with the probabilities P₀ and 1 ? P₀, respectively, then the root node is almost a sure MIN or a sure LOSS, depending on whether P₀ is higher or lower than some fixed-point probability $\text{P1(d)}$. When the terminal positions are assigned continuous real values, the minimax value of the root node converges rapidly to a unique predetermined value $\text{v1}$, which is the $\text{(1 ? P1)-}\text{fractile}$ of the terminal distribution.Exploiting these properties we show that a game with WIN-LOSS terminals can be solved by examining, on the average, $\text{O}\text{[(d)}{}^{\mathrm{<mtext>h</mtext><mtext>2</mtext>}}\text{]}$ terminal positions if positions if $\text{P}{}_0\text{≠ P1}$ and $\text{O}\text{[(}\text{P1}\text{(1 ? P1)}\text{)}{}^{\mathrm{h}}\text{]}$ positions if $\text{P}{}_0\text{= P1}$, the former performance being optimal for all search algorithms. We further show that a game with continuous terminal values can be evaluated by examining an average of $\text{O}\text{[(}\text{P1}\text{(1 ? P1)}\text{)}{}^{\mathrm{h}}\text{]}$ positions, and that this is a lower bound for all directional algorithms. Games with discrete terminal values can, in almost all cases, be evaluated by examining an average of $\text{O}\text{[(d)}{}^{\mathrm{<mtext>h</mtext><mtext>2</mtext>}}\text{]}$ terminal positions. This performance is optimal and is also achieved by the ALPHA-BETA procedure.     

Asymptotic properties of solutions to discrete Coulomb equations

We construct two finite-difference models for the Coulomb differential equation which arises in the quantum mechanics analysis of the scattering of two charged point particles. These difference equations correspond to the standard and Mickens-Ramadhani schemes for the Coulomb equation. Our major goal is to determine the first two terms in the asymptotic solutions and compare them to the corresponding solutions of the Coulomb differential equation. In particular, the form of the anomalous phase term is examined.