Quadtree-structured linear prediction models for image sequenceprocessing

A summary is presented of a study on two-dimensional linear prediction models for image sequence processing and its application to change detection and scene coding. The study focused on two-dimensional joint process modeling of interframe relationships, the derivation of computationally efficient matching algorithms, and the implementation of a block-adaptive interframe predictor for use in interframe predictive coding and change detection. In the approach presented, the spatial nonstationarity is handled by an underlying quadtree segmentation structure. A maximum-likelihood criterion and a simpler minimum-variance criterion are discussed as detection and segmentation rules. The results of this research indicate that a constrained joint process model involving only a single gain parameter and a shift parameter is the best tradeoff between performance and computational complexity     

On deterministic chaos in software reliability growth models

Software reliability growth models attempt to forecast the future reliability of a software system, based on observations of the historical occurrences of failures. This allows management to estimate the failure rate of the system in field use, and to set release criteria based on these forecasts. However, the current software reliability growth models have never proven to be accurate enough for widespread industry use. One possible reason is that the model forms themselves may not accurately capture the underlying process of fault injection in software; it has been suggested that fault injection is better modeled as a chaotic process rather than a random one. This possibility, while intriguing, has not yet been evaluated in large-scale, modern software reliability growth datasets.We report on an analysis of four software reliability growth datasets, including ones drawn from the Android and Mozilla open-source software communities. These are the four largest software reliability growth datasets we are aware of in the public domain, ranging from 1200 to over 86,000 observations. We employ the methods of nonlinear time series analysis to test for chaotic behavior in these time series; we find that three of the four do show evidence of such behavior (specifically, a multifractal attractor). Finally, we compare a deterministic time series forecasting algorithm against a statistical one on both datasets, to evaluate whether exploiting the apparent chaotic behavior might lead to more accurate reliability forecasts.     

Validity and reliability of evaluation procedures in comparative studies of effort prediction models

We have in previous studies reported our findings and concern about the reliability and validity of the evaluation procedures used in comparative studies on competing effort prediction models. In particular, we have raised concerns about the use of accuracy statistics to rank and select models. Our concern is strengthened by the observed lack of consistent findings. This study offers more insights into the causes of conclusion instability by elaborating on the findings of our previous work concerning the reliability and validity of the evaluation procedures. We show that model selection based on the accuracy statistics MMRE, MMER, MBRE, and MIBRE contribute to conclusion instability as well as selection of inferior models. We argue and show that the evaluation procedure must include an evaluation of whether the functional form of the prediction model makes sense to better prevent selection of inferior models.     

On linear models for nonlinear systems

Best linear time-invariant (LTI) approximations are analysed for several interesting classes of discrete nonlinear time-invariant systems. These include nonlinear finite impulse response systems and a class of nonsmooth systems called bi-gain systems. The Fréchet derivative of a smooth nonlinear system is studied as a potential good LTI model candidate. The Fréchet derivative is determined for nonlinear finite memory systems and for a class of Wiener systems. Most of the concrete results are derived in an ?_∞ signal setting. Applications to linear controller design, to identification of linear models and to estimation of the size of the unmodelled dynamics are discussed.     

Explanation and reliability of prediction models: the case of breast cancer recurrence

In this paper, we describe the first practical application of two methods, which bridge the gap between the non-expert user and machine learning models. The first is a method for explaining classifiers’ predictions, which provides the user with additional information about the decision-making process of a classifier. The second is a reliability estimation methodology for regression predictions, which helps the users to decide to what extent to trust a particular prediction. Both methods are successfully applied to a novel breast cancer recurrence prediction data set and the results are evaluated by expert oncologists.     

On the stability issues of linear Takagi-Sugeno fuzzy models

Stability issues of linear Takagi-Sugeno (TS) fuzzy models (1985, 1992) are investigated. We first propose a systematic way of searching for a common matrix, which, in turn, is related to stability for N subsystems that are under a pairwise commutative assumption. The robustness issue under uncertainty in each subsystem is then considered. We then show that the pairwise commutative assumption can, in fact, be relaxed by a similar approach as that for uncertainty. The result is applicable to a rather broad class of TS models, which are nonHurwitz and/or nonpairwise commutative     

The reliability of analogy-based prediction

This paper is focused on choosing a sufficient number of runs of a coupling Markov chain that makes it possible to generate, with a high confidence level, hypotheses such that at least one of them is inserted into any test example with high probability of positive prediction. The proposed technique is based on the Vapnik–Chervonenkis resampling method.     

Using neural networks in reliability prediction

It is shown that neural network reliability growth models have a significant advantage over analytic models in that they require only failure history as input and not assumptions about either the development environment or external parameters. Using the failure history, the neural-network model automatically develops its own internal model of the failure process and predicts future failures. Because it adjusts model complexity to match the complexity of the failure history, it can be more accurate than some commonly used analytic models. Results with actual testing and debugging data which suggest that neural-network models are better at endpoint predictions than analytic models are presented     

On the dataset shift problem in software engineering prediction models

A core assumption of any prediction model is that test data distribution does not differ from training data distribution. Prediction models used in software engineering are no exception. In reality, this assumption can be violated in many ways resulting in inconsistent and non-transferrable observations across different cases. The goal of this paper is to explain the phenomena of conclusion instability through the dataset shift concept from software effort and fault prediction perspective. Different types of dataset shift are explained with examples from software engineering, and techniques for addressing associated problems are discussed. While dataset shifts in the form of sample selection bias and imbalanced data are well-known in software engineering research, understanding other types is relevant for possible interpretations of the non-transferable results across different sites and studies. Software engineering community should be aware of and account for the dataset shift related issues when evaluating the validity of research outcomes.     

Recalibrating software reliability models

There is no universally applicable software reliability growth model which can be trusted to give accurate predictions of reliability in all circumstances. A technique of analyzing predictive accuracy called the u-plot allows a user to estimate the relationship between the predicted reliability and the true reliability. It is shown how this can be used to improve reliability predictions in a very general way by a process of recalibration. Simulation results show that the technique gives improved reliability predictions in a large proportion of cases. However, a user does not need to trust the efficacy of recalibration, since the new reliability estimates produced by the technique are truly predictive and their accuracy in a particular application can be judged using the earlier methods. The generality of this approach suggests its use whenever a software reliability model is used. Indeed, although this work arose from the need to address the poor performance of software reliability models, it is likely to have applicability in other areas such as reliability growth modeling for hardware     

On optimal approximation of high-order linear systems by low-order models

An engineering solution to the problem of approximating a high-order linear system by a low-order model is suggested, which consists in formulating a quadratic error functional as a basic tool for the construction of a physically transparent and practically meaningful approximation criterion. For various different modes of operation analytical expressions for the evaluation of the functional as well as the gradients with respect to the unknown model parameters are given, which allows known gradient techniques to be applied for finding the optimal model parameters. In addition a representation for a general multivariate model, using only the minimal number of unknown parameters, is given.     

Robust feedforward and recurrent neural network based dynamic weighted combination models for software reliability prediction

Traditional parametric software reliability growth models (SRGMs) are based on some assumptions or distributions and none such single model can produce accurate prediction results in all circumstances. Non-parametric models like the artificial neural network (ANN) based models can predict software reliability based on only fault history data without any assumptions. In this paper, initially we propose a robust feedforward neural network (FFNN) based dynamic weighted combination model (PFFNNDWCM) for software reliability prediction. Four well-known traditional SRGMs are combined based on the dynamically evaluated weights determined by the learning algorithm of the proposed FFNN. Based on this proposed FFNN architecture, we also propose a robust recurrent neural network (RNN) based dynamic weighted combination model (PRNNDWCM) to predict the software reliability more justifiably. A real-coded genetic algorithm (GA) is proposed to train the ANNs. Predictability of the proposed models are compared with the existing ANN based software reliability models through three real software failure data sets. We also compare the performances of the proposed models with the models that can be developed by combining three or two of the four SRGMs. Comparative studies demonstrate that the PFFNNDWCM and PRNNDWCM present fairly accurate fitting and predictive capability than the other existing ANN based models. Numerical and graphical explanations show that PRNNDWCM is promising for software reliability prediction since its fitting and prediction error is much less relative to the PFFNNDWCM.     

On the slow convergence of EM and VBEM in low-noise linear models

We analyze convergence of the expectation maximization (EM) and variational Bayes EM (VBEM) schemes for parameter estimation in noisy linear models. The analysis shows that both schemes are inefficient in the low-noise limit. The linear model with additive noise includes as special cases independent component analysis, probabilistic principal component analysis, factor analysis, and Kalman filtering. Hence, the results are relevant for many practical applications.     

不完全数据的软件可靠性预测

针对现有处理不完全数据的填充方法,对数据集引入新的噪声这一问题,提出一种基于最大间隔理论的预测模型,直接使用含缺失特征的样本进行预测,通过实验验证该模型优于常用的填充模型。     

On discrete models and immunological algorithms for protein structure prediction

Discrete models for protein structure prediction embed the protein amino acid sequence into a discrete spatial structure, usually a lattice, where an optimal tertiary structure is predicted on the basis of simple assumptions relating to the hydrophobic–hydrophilic character of amino acids in the sequence and to relevant interactions for free energy minimization. While the prediction problem is known to be NP complete even in the simple setting of Dill’s model with a 2D-lattice, a variety of bio-inspired algorithms for this problem have been proposed in the literature. Immunological algorithms are inspired by the kind of optimization that immune systems perform when identifying and promoting the replication of the most effective antibodies against given antigens. A quick, state-of-the-art survey of discrete models and immunological algorithms for protein structure prediction is presented in this paper, and the main design and performance features of an immunological algorithm for this problem are illustrated in a tutorial fashion.     

Testing linear independence in linear models with interval-valued data

Testing methods are introduced in order to determine whether there is some ‘linear’ relationship between imprecise predictor and response variables in a regression analysis. The variables are assumed to be interval-valued. Within this context, the variables are formalized as compact convex random sets, and an interval arithmetic-based linear model is considered. Then, a suitable equivalence for the hypothesis of linear independence in this model is obtained in terms of the mid-spread representations of the interval-valued variables. That is, in terms of some moments of random variables. Methods are constructed to test this equivalent hypothesis; in particular, the one based on bootstrap techniques will be applicable in a wide setting. The methodology is illustrated by means of a real-life example, and some simulation studies are considered to compare techniques in this framework.     

Time-dependent reliability prediction using transfer learning

Structural and Multidisciplinary Optimization - One of the key challenges in reliability estimation is the acquisition of failure information especially under real-life scenarios or computationally...     

Estimating reliability in proportional odds ratio models

In this paper, we are mainly interested in inference on the reliability coefficient, R=P(X<Y), in proportional odds ratio models based on the new family of tilted survival functions introduced by Marshall and Olkin [Marshall, A.W., Olkin, I., 1997. A new method for adding a parameter to a family of distributions with application to the exponential and Weibull families. Biometrika 84 (3), 641-652]. We also present some results on stochastic comparison between the survival distribution functions. Asymptotic and various bootstrap confidence intervals of R are investigated. The performance of asymptotic and bootstrap confidence intervals is studied through a simulation. A numerical example based on real-life data is presented to illustrate the implementation of the proposed procedure.     

On the robustness of the dominant eigenvalue of dynamic linear econometric models

Uebe and Fischer [Comput. Ops. Res., 1, 313–339 (1974)] have investigated heuristically the sensitivity of the dominant eigenvalue of a number of well-known econometric models with respect to the deletion of small model coefficients. In this comment we demonstrate that the observed robustness can be explained to be the result of the presence of dynamic model equations of a specific type. Therefore we have evaluated the elasticities pertaining to the relevant eigenvalues with respect to the model coefficients. We also argue that the mentioned type of equations causes many econometric models to have a dominant eigenvalue close to unity. In the light of our findings we critically examine the interpretation usually attached to such dominant eigenvalues.     

Collinearity detection in linear regression models

Multicollinearity can seriously affect least-squares parameter estimates. Many methods have been suggested to determine those parameters most involved. This paper, beginning with the contributions of Belsley, Kuh, and Welsch (1980) and Belsley (1991), forges a new direction. A decomposition of the variable space allows the near dependencies to be isolated in one sub-space. And this, in turn, allows a corresponding decomposition of the main statistics, as well as a new one proposed here, to provide better information on the structure of the collinear relations.