Results 1  10
of
24
When Networks Disagree: Ensemble Methods for Hybrid Neural Networks
, 1993
"... This paper presents a general theoretical framework for ensemble methods of constructing significantly improved regression estimates. Given a population of regression estimators, we construct a hybrid estimator which is as good or better in the MSE sense than any estimator in the population. We argu ..."
Abstract

Cited by 290 (2 self)
 Add to MetaCart
This paper presents a general theoretical framework for ensemble methods of constructing significantly improved regression estimates. Given a population of regression estimators, we construct a hybrid estimator which is as good or better in the MSE sense than any estimator in the population. We argue that the ensemble method presented has several properties: 1) It efficiently uses all the networks of a population  none of the networks need be discarded. 2) It efficiently uses all the available data for training without overfitting. 3) It inherently performs regularization by smoothing in functional space which helps to avoid overfitting. 4) It utilizes local minima to construct improved estimates whereas other neural network algorithms are hindered by local minima. 5) It is ideally suited for parallel computation. 6) It leads to a very useful and natural measure of the number of distinct estimators in a population. 7) The optimal parameters of the ensemble estimator are given in clo...
Improving Regression Estimation: Averaging Methods for Variance Reduction with Extensions to General Convex Measure Optimization
, 1993
"... ..."
Bias and Variance of Validation Methods for Function Approximation Neural Networks Under Conditions of Sparse Data
 IEEE Transactions on Systems, Man, and Cybernetics, Part C
, 1998
"... Neural networks must be constructed and validated with strong empirical dependence, which is difficult under conditions of sparse data. This paper examines the most common methods of neural network validation along with several general validation methods from the statistical resampling literature ..."
Abstract

Cited by 11 (6 self)
 Add to MetaCart
Neural networks must be constructed and validated with strong empirical dependence, which is difficult under conditions of sparse data. This paper examines the most common methods of neural network validation along with several general validation methods from the statistical resampling literature as applied to function approximation networks with small sample sizes. It is shown that an increase in computation, necessary for the statistical resampling methods, produces networks that perform better than those constructed in the traditional manner. The statistical resampling methods also result in lower variance of validation, however some of the methods are biased in estimating network error. 1. INTRODUCTION To be beneficial, system models must be validated to assure the users that the model emulates the actual system in the desired manner. This is especially true of empirical models, such as neural network and statistical models, which rely primarily on observed data rather th...
Modelling Heterogeneity in Cetacean Surveys
, 2000
"... Methods for improving estimation of cetacean abundance from line transect and mark recapture surveys are proposed. Using either generalized linear or generalized additive models (GLMs or GAMs), two approaches are suggested which allow heterogeneity in the spatial distribution of cetaceans to be mode ..."
Abstract

Cited by 4 (0 self)
 Add to MetaCart
Methods for improving estimation of cetacean abundance from line transect and mark recapture surveys are proposed. Using either generalized linear or generalized additive models (GLMs or GAMs), two approaches are suggested which allow heterogeneity in the spatial distribution of cetaceans to be modelled from standard line transect data. In the first approach, the transect lines are divided into smaller discrete units, and the expected number of detections in each unit is modelled using explanatory spatial covariates. In the second approach, the response is derived from the observed waiting times (or distances) between detections. Fitting this model within the usual GLM or GAM framework would require restrictive assumptions, therefore an iterative procedure is formulated which enables a realistic model to be fitted. Alternatively, it is shown how this approach can be framed as a point process model, and it is suggested how the likelihood for the observed alongtrackline distances could be maximized. The methods are illustrated using line transect data from a survey of Antarctic minke whales. A surface representing the variation in density throughout the survey region is obtained, from which abundance may be estimated by numerical integration. It is also
Applying a robust heteroscedastic probabilistic neural network to analog fault detection and classification
 IEEE Transactions on
, 2000
"... Abstract—The problem of distinguishing and classifying the responses of analog integrated circuits containing catastrophic faults has aroused recent interest. The problem is made more difficult when parametric variations are taken into account. Hence, statistical methods and techniques such as neura ..."
Abstract

Cited by 3 (0 self)
 Add to MetaCart
Abstract—The problem of distinguishing and classifying the responses of analog integrated circuits containing catastrophic faults has aroused recent interest. The problem is made more difficult when parametric variations are taken into account. Hence, statistical methods and techniques such as neural networks have been employed to automate classification. The major drawback to such techniques has been the implicit assumption that the variances of the responses of faulty circuits have been the same as each other and the same as that of the faultfree circuit. This assumption can be shown to be false. Neural networks, moreover, have proved to be slow. This paper describes a new neural network structure that clusters responses assuming different means and variances. Sophisticated statistical techniques are employed to handle situations where the variance tends to zero, such as happens with a fault that causes a response to be stuck at a supply rail. Two example circuits are used to show that this technique is significantly more accurate than other classification methods. A set of responses can be classified in the order of 1 s. Index Terms—Automatic test pattern generation (ATPG) testing, faultdiagnosis, quiescent supply current (IDDQ), mixedsignal_test. I.
Fault detection and classification in analogue integrated circuits using robust heteroscedastic probabilistic neural networks
 4th IEEE International Mixed Signal Testing Workshop, The Hague, The
, 1998
"... A new neural network method has been used to distinguish faulty from faultfree circuit responses. This technique is significantly more accurate than other classification methods. A set of responses can be classified in the order of 1 second. 1. ..."
Abstract

Cited by 2 (0 self)
 Add to MetaCart
A new neural network method has been used to distinguish faulty from faultfree circuit responses. This technique is significantly more accurate than other classification methods. A set of responses can be classified in the order of 1 second. 1.
The Variance of the Mutual Information Estimator
, 1997
"... In the case of two signals with independent pairs of observations (xn ; yn) a statistic to estimate the variance of the mutual information estimator has been derived earlier. We present such a statistic for dependent pairs. To derive this statistic it is necessary to avail of a reliable statistic t ..."
Abstract

Cited by 2 (0 self)
 Add to MetaCart
In the case of two signals with independent pairs of observations (xn ; yn) a statistic to estimate the variance of the mutual information estimator has been derived earlier. We present such a statistic for dependent pairs. To derive this statistic it is necessary to avail of a reliable statistic to estimate the variance of the sample mean in case of dependent observations. We derive and discuss this statistic and a statistic to estimate the variance of the mutual information estimator. These statistics are verified by simulations. Key words: sample mean, variance, statistic, mutual information, entropy, likelihood, dependent samples. 1 Introduction To estimate timedelays between recordings of electroencephalogram (EEG) signals Mars and van Arragon introduced a method based on maximum mutual information [13]. The method was modified and extended [15, 16]. Mutual information is used to measure the dependence of nonlinearly dependent random variables and widely applied to stochasti...
Parametric Distance Metrics Vs. Nonparametric Neural Networks For Estimating Road Travel Distances
, 1994
"... The actual distance betveen tvo cities is the length of the shortest road connecting them. Measuring and storing the actual distance between any two points of a region is often not feasible and it is a common practice to estimate it. The usual approach is to use theoretical distance metrics vhich ..."
Abstract

Cited by 1 (0 self)
 Add to MetaCart
The actual distance betveen tvo cities is the length of the shortest road connecting them. Measuring and storing the actual distance between any two points of a region is often not feasible and it is a common practice to estimate it. The usual approach is to use theoretical distance metrics vhich are parameterized functions of the coordinates of the points. We propose to use nonparametric approaches using neural netvorks for estimating actual distances. We consider multilayer percepttons trained vith the backpropagation rule and regression neural netvorks implementing nonparametric regression using Gaussian kernels.
Berkeley Earth Temperature Averaging Process
"... A new mathematical framework is presented for producing maps and largescale averages of temperature changes from weather station data for the purposes of climate analysis. This allows one to include short and discontinuous temperature records, so that nearly all temperature data can be used. The fr ..."
Abstract

Cited by 1 (0 self)
 Add to MetaCart
A new mathematical framework is presented for producing maps and largescale averages of temperature changes from weather station data for the purposes of climate analysis. This allows one to include short and discontinuous temperature records, so that nearly all temperature data can be used. The framework contains a weighting process that assesses the quality and consistency of a spatial network of temperature stations as an integral part of the averaging process. This permits data with varying levels of quality to be used without compromising the accuracy of the resulting reconstructions. Lastly, the process presented here is extensible to spatial networks of arbitrary density (or locally varying density) while maintaining the expected spatial relationships. In this paper, this framework is applied to the Global Historical Climatology Network land temperature dataset to present a new global land temperature reconstruction from 1800 to present with error uncertainties that include many key effects. In so doing, we find that the global land mean temperature has increased by 0.911 ± 0.042 C since the 1950s (95 % confidence for statistical and spatial uncertainties). This change is consistent with global landsurface warming results previously reported, but with reduced uncertainty. 2
Classifier Evaluation and the Use of Algorithmic Classifiers With Expert System Classifiers
, 1993
"... Classification is an important task in the realworld. It is important to develop techniques for the comparison of classifiers. This has to be done statistically and in terms of welldefined criteria. This is a major goal of this thesis. This thesis discusses such techniques specifically in the cont ..."
Abstract
 Add to MetaCart
Classification is an important task in the realworld. It is important to develop techniques for the comparison of classifiers. This has to be done statistically and in terms of welldefined criteria. This is a major goal of this thesis. This thesis discusses such techniques specifically in the context of comparing algorithmic classifiers (algorithms) and expert system classifiers (expert systems). The basic criterion used in this thesis is L, the weighted loss of misclassification. L is a wellestablished criterion in the statistical and patternrecognition literature. A relatively unknown criterion by Brennan, kappa base ( b ) that adjusts for agreement attainable due to chance, is introduced from the biomedical and psychological literature. b is extended to b;w for the case when the misclassifications are of varying degrees of severity. The importance of using statistical techniques such as hypothesis testing and confidence intervals is stressed. The use of crossvalidation, th...