摘要:

Artificial neural networks are computer algorithms or computer programs derived in part from attempts to model the activity of nerve cells. They have been applied to pattern recognition, classification, and optimization problems in the physical and chemical sciences, as well as in other fields. We introduce the principles of the multilayer feedforward network that is among the most commonly used neural networks in practical problems. The relevance of neural network models for the applied statistician is considered using a time series prediction problem as an example. The multilayer feedforward neural network uses a nonlinear function of the predictors to obtain predictions for future time series values. We illustrate the considerations involved in specifying a neural network model and evaluate the accuracy of neural network models relative to the accuracy obtained using other computer-intensive, nonmodel-based techniques.

ISSN:0040-1706    

DOI:10.1080/00401706.1996.10484497

出版商:Taylor & Francis Group    

年代:1996

数据来源: Taylor

摘要:

In this discussion, we give a short overview of some of the uses of artificial neural networks in practice. This is not intended to be an exhaustive list, but it includes some widely used industrial examples with which we are familiar. From our experience, we also attempt to point out why they are used and what the relative advantages and disadvantages of using a neural network over a traditional statistical method might be.

ISSN:0040-1706    

DOI:10.1080/00401706.1996.10484498

出版商:Taylor & Francis Group    

年代:1996

数据来源: Taylor

ISSN:0040-1706    

DOI:10.1080/00401706.1996.10484499

出版商:Taylor & Francis Group    

年代:1996

数据来源: Taylor

ISSN:0040-1706    

DOI:10.1080/00401706.1996.10484500

出版商:Taylor & Francis Group    

年代:1996

数据来源: Taylor

摘要:

Calibration consists of using a fitted regression line to estimate the value of an unobserved independent variablexcorresponding to an observed dependent variabley. To construct a confidence interval for a singlex, Eisenhart introduced a procedure that consists of inverting prediction intervals around the regression line. Numerous other inference procedures have been proposed for multiple-use calibration, in which a single fitted regression line is used repeatedly to estimate manyx's, We provide a synthesis of this literature and offer some numerical comparisons. We also attempt to motivate the use of various criteria based on the particular points of view of the various parties involved in determining the calibration or using the results. In addition, we derive probability expressions for computing exact simultaneous prediction intervals that enable the construction of tighter limits than are currently available based on that criterion.

ISSN:0040-1706    

DOI:10.1080/00401706.1996.10484501

出版商:Taylor & Francis Group    

年代:1996

数据来源: Taylor

摘要:

Methodology is proposed for the design of sequential methods when data are obtained by gauging articles into groups, Exact expressions are obtained for the operating characteristics and average sampling number of Wald sequential probability ratio tests and for the average run length of cumulative sum (CUSUM) schemes based on grouped data. Step-by-step design algorithms are provided to assist the practitioner. The methodology is illustrated assuming a normal process with known standard deviation in which we wish to detect shifts in the mean. An example from a progressive die operation is presented. The methods proposed are simple to implement and are an economical alternative to variables-data-based sequential sampling plans and CUSUM control charts.

ISSN:0040-1706    

DOI:10.1080/00401706.1996.10484502

出版商:Taylor & Francis Group    

年代:1996

数据来源: Taylor

摘要:

In the interpretation of the geometry of second-order response surface models, standard errors and confidence intervals for the eigenvalues of the second-order coefficient matrix play an important role. In this article, we propose a new method for estimating the standard errors, and hence approximate confidence intervals, of these eigenvalues. The method is simple in both concept and execution. It involves the refitting of a full quadratic model after rotating the coordinate system to coincide with the canonical axes. The estimated standard errors of the pure quadratic terms from this refitting are then used as approximate standard errors of the eigenvalues. Because this approach is based on the canonical form, it is geometrically intuitive and easily taught. Our method is intended as a way for practitioners to get quick estimates of the standard errors of the eigenvalues. In our justification of the approach, we show that it is equivalent to using the delta method proposed by Carter, Chinchilli, and Campbell.

ISSN:0040-1706    

DOI:10.1080/00401706.1996.10484503

出版商:Taylor & Francis Group    

年代:1996

数据来源: Taylor

摘要:

This article considers the situation in which a system consists ofkcomponents and a defect in any component causes a system malfunction. When a system malfunction occurs, test procedures restrict the cause to some subset of the Ic components. When that subset consists of more than one component, this phenomenon is termedmasking. Typically, masking introduces two types of problems. First, it is desirable to estimate the “diagnostic probability”—that is, the probability, given a specified malfunctioning subset, that each of the masked components is the defective one. Second, when a set of historical data contains masked information, one would like to use this information to estimate the defect probability of each individual component type. The article discusses these problems in detail and derives two-stage procedures for estimation and inference.

ISSN:0040-1706    

DOI:10.1080/00401706.1996.10484504

出版商:Taylor & Francis Group    

年代:1996

数据来源: Taylor

摘要:

The power law process (PLP) is often used to model failure data from repairable systems, and its adequacy is tested with goodness-of-fit tests such as the Cramér–von Mises test. No single test, however, can be sensitive to all possible deviations from the PLP. Exact tests that are powerful against some departures from the PLP of particular interest to practitioners are derived heuristically, the computation of theirpvalues is described, and their use is illustrated with examples. One test is a modified version of the Laplace test for trend, another a modification of Greenwood's test for clustering. It is shown that these tests can be derived as score tests, which ensures that they are powerful for large samples.

ISSN:0040-1706    

DOI:10.1080/00401706.1996.10484505

出版商:Taylor & Francis Group    

年代:1996

数据来源: Taylor

摘要:

Histogram and scatterplot models are often required for statistical inference. In the field of petroleum engineering, stochastic simulation algorithms require, among other statistics, a model for the histogram of the petrophysical attribute (porosity/permeability) being simulated. Often this model is taken to be the deciustered distribution of the sample data. When there are many data (say, greater than l,000), this histogram may be reasonably informed. Most often, however, the sample histogram shows multiple sawtoothlike spikes that are not representative of the entire population; the sample histogram must be smoothed. A simulated annealing-based procedure is proposed for smoothing one-variable (univariate) histograms and two-variable scatterplots (bivariate histograms). The smoothed histograms are constrained to the sample mean, variance, specified quantiles, and a measure of smoothness. In the bivariate case, the distribution must be consistent with both marginal histograms and can be additionally constrained to a linear correlation coefficient. Several examples with real reservoir data are presented.

ISSN:0040-1706    

DOI:10.1080/00401706.1996.10484506

出版商:Taylor & Francis Group    

年代:1996

数据来源: Taylor