Variance-covariance based weighing for neural network ensembles

Neural network (NN) is a popular artificial intelligence technique for solving complicated problems due to their inherent capabilities. However generalization in NN can be harmed by a number of factors including parameter's initialization, inappropriate network topology and setting parameters of the training process itself. Forecast combinations of NN models have the potential for improved generalization and lower training time. A weighted averaging based on Variance-Covariance method that assigns greater weight to the forecasts producing lower error, instead of equal weights is practiced in this paper. While implementing the method, combination of forecasts is done with all candidate models in one experiment and with the best selected models in another experiment. It is observed during the empirical analysis that forecasting accuracy is improved by combining the best individual NN models. Another finding of this study is that reducing the number of NN models increases the diversity and, hence, accuracy.

Genetic correlation between resting metabolic rate and exploratory behaviour in deer mice (Peromyscus maniculatus)

According to the ‘pace-of-life’ syndrome hypothesis, differences in resting metabolic rate (RMR) should be genetically associated with exploratory behaviour. A large number of studies reported significant heritability for both RMR and exploratory behaviour, but the genetic correlation between the two has yet to be documented. We used a quantitative genetic approach to decompose the phenotypic (co)variance of several metabolic and behavioural measures into components of additive genetic, common environment and permanent environment variance in captive deer mice. We found significant additive genetic variance for two mass-independent metabolic measures (RMR and the average metabolic rate throughout the respirometry run) and two behavioural measures (time spent in centre and distance moved in a novel environment). We also detected positive additive genetic correlation between mass-independent RMR and distance moved (rA = 0.78 ± 0.23). Our results suggest that RMR and exploratory behaviour are functionally integrated traits in deer mice, providing empirical support for one of the connections within the pace-of-life syndrome hypothesis.

The effects of clustered data on standard error estimates in covariance structure analysis : a field data application

Purpose – The purpose of this article is to present an empirical analysis of complex sample data with regard to the biasing effect of non-independence of observations on standard error parameter estimates. Using field data structured in the form of repeated measurements it is to be shown, in a two-factor confirmatory factor analysis model, how the bias in SE can be derived when the non-independence is ignored.

Design/methodology/approach – Three estimation procedures are compared: normal asymptotic theory (maximum likelihood); non-parametric standard error estimation (naïve bootstrap); and sandwich (robust covariance matrix) estimation (pseudo-maximum likelihood).

Findings – The study reveals that, when using either normal asymptotic theory or non-parametric standard error estimation, the SE bias produced by the non-independence of observations can be noteworthy.

Research limitations/implications – Considering the methodological constraints in employing field data, the three analyses examined must be interpreted independently and as a result taxonomic generalisations are limited. However, the study still provides “case study” evidence suggesting the existence of the relationship between non-independence of observations and standard error bias estimates.

Originality/value – Given the increasing popularity of structural equation models in the social sciences and in particular in the marketing discipline, the paper provides a theoretical and practical insight into how to treat repeated measures and clustered data in general, adding to previous methodological research. Some conclusions and suggestions for researchers who make use of partial least squares modelling are also drawn.

Genetic and environmental influences on variation in balance performance among female twin pairs aged 21-82 years

Genetic and environmental influences on variation in balance performance were measured in 93 monozygous and 83 dizygous female twin pairs aged 21–82 years (mean age, 50.5 years) in Melbourne, Australia, between 1999 and 2003. The authors administered clinical (Lord's Balance Test and Step Test) and laboratory tests of static and dynamic balance from the Chattecx Balance System with and without distractor tasks. The authors conducted factor analysis and estimated genetic and environmental variance components and heritability (defined as additive genetic variance as a proportion of all variance, after adjustment for age) using a multivariate normal model with the statistical package FISHER. Three factors were identified and adjusted for age. Heritability was 46% (standard error (SE), 9) for the "sensory balance tests" factor and 30% (SE, 9) for the "static and dynamic perturbations" factor. For both factors, the remaining variance was attributed to unique environmental effects. There was no evidence that genetic factors influenced variation in the "dynamic weight shift tests" factor, with environmental effects shared by twins accounting for 38% (SE, 7) of variance. Neither genetic nor environmental proportions of variance differed significantly between twin subgroups by age (≤50/>50 years). An age-related decline in performance measures was found across the whole sample. These results imply that balance impairments may have a heritable element.

Modelling for injection moulding process based on principal component regression method

Determination of the optimal operating condition for moulding process has been of special interest for many researchers. To determine the optimal setting, one has to derive the model of injection moulding process first which is able to map the relationship between the input process control factors and output responses. One of most popular modeling techniques is the linear least square regression due to its effectiveness and completeness. However, the least square regression was found to be very sensitive to the outliers and failed to provide a reliable model if the control variables are highly related with each other. To address this problem, a new modeling method based on principal component regression was proposed in this paper. The distinguished feature of our proposed method is it does not only consider the variance of covariance matrix of control variables but also consider the correlation coefficient between control variables and target variables to be optimised. Such a modelling method has been implemented into a commercial optimisation software and field test results demonstrated the performance of the proposed modelling method.

Mitochondrial DNA diversity of the Asian moon scallop, Amusium pleuronectes (Pectinidae), in Thailand

Sequence variation of the mitochondrial DNA 16S rRNA region of the Asian moon scallop, Amusium pleuronectes, was surveyed in seven populations along the coast of Thailand. A total of 16 unique haplotypes were detected among 174 individuals with a total 27 variable sites out of 534 bp sequenced. The mitochondrial haplotypes grouped into two distinct arrays (estimated to differ by about 2.62% to 2.99% nucleotide divergence) that characterized samples collected from the Gulf of Thailand versus the Andaman Sea. Low levels of intrapopulation variation were observed, while in contrast, significant divergence was observed between populations from the Gulf of Thailand and Andaman Sea. Results of AMOVA reveal a high F ST value (0.765) and showed that the majority of the total genetic variance (76.03%) occurred among groups (i.e., Andaman Sea and the Gulf of Thailand) and little among populations within the group (0.52%) and within populations (23.45%). The genetic differentiation between the populations recorded in the present study is similar to that observed in a variety of marine species in the Indo-Pacific. The implications of the findings for management of A. pleuronectes genetic resources in Thailand are discussed.

Genome - wide association study identifies novel breast cancer susceptibility loci

Breast cancer exhibits familial aggregation, consistent with variation in genetic susceptibility to the disease. Known susceptibility genes account for less than 25% of the familial risk of breast cancer, and the residual genetic variance is likely to be due to variants conferring more moderate risks. To identify further susceptibility alleles, we conducted a two-stage genome-wide association study in 4,398 breast cancer cases and 4,316 controls, followed by a third stage in which 30 single nucleotide polymorphisms (SNPs) were tested for confirmation in 21,860 cases and 22,578 controls from 22 studies. We used 227,876 SNPs that were estimated to correlate with 77% of known common SNPs in Europeans at r² > 0.5. SNPs in five novel independent loci exhibited strong and consistent evidence of association with breast cancer (P < 10^-7). Four of these contain plausible causative genes (FGFR2, TNRC9, MAP3K1 and LSP1). At the second stage, 1,792 SNPs were significant at the P < 0.05 level compared with an estimated 1,343 that would be expected by chance, indicating that many additional common susceptibility alleles may be identifiable by this approach.

Population genetics of the spotted seahorse (Hippocampus kuda) in Thai waters : implications for conservation

A population genetics approach was used to investigate the genetic diversity of the spotted seahorse (Hippocampus kuda) in Thai waters; specifically, the degree of genetic differentiation and species evolution was inferred from sequence analysis of 353 bp of the mitochondrial (mt)DNA control region. The data were then used to identify discrete populations in Thai waters for effective conservation and management. Spotted seahorses were collected from 4 regions on the east and west coasts of the Gulf of Thailand and a geographically separated region in the Andaman Sea. Of the 101 mtDNA sequences analyzed, 7 haplotypes were identified, 5 of which were shared among individuals from the east and west coasts of the Gulf of Thailand. The remaining haplotypes were restricted to individuals from the Andaman Sea. Nucleotide and haplotype diversities were similar within the Gulf of Thailand samples, whereas diversity was lower in the Andaman Sea sample. Genetic differentiation appeared between pairs of samples from the Gulf of Thailand and Andaman Sea (FST, p < 0.0001). A large genetic variance appeared among the 2 population groups (94.46%, ΦCT = 0.94464, p < 0.01). A Neighbor-joining tree indicated that individuals from the Gulf of Thailand and Andaman Sea formed 2 phylogenetically distinct groups, which were segregated into different population-based clades. While results reported here indicate that populations from the Gulf of Thailand and Andaman Sea should be treated as separate conservation units, a larger sample size from the Andaman Sea is required to confirm this genetic partitioning and low level of diversity observed in the present study.

Automatic parameters selection for eigenfaces

In this paper, we investigate the parameters selection for Eigenfaces. Our focus is on the eigenvectors and threshold selection issues. We will propose a systematic approach in selecting the eigenvectors based on relative errors of the eigenvalues for the covariance matrix. In addition, we have proposed a method for selecting the classification threshold that utilizes the information obtained from the training data set. Experimentation was conducted on two benchmark face databases, ORL and AMP, with results indicating that the proposed automatic eigenvectors and threshold selection methods produce better recognition performance in terms of precision and recall rates. Furthermore, we show that the eigenvector selection method outperforms energy and stretching dimension methods in terms of selected number of eigenvectors and computation cost.

Boosting performance for 2D linear discriminant analysis via regression

Two Dimensional Linear Discriminant Analysis (2DLDA) has received much interest in recent years. However, 2DLDA could make pairwise distances between any two classes become significantly unbalanced, which may affect its performance. Moreover 2DLDA could also suffer from the small sample size problem. Based on these observations, we propose two novel algorithms called Regularized 2DLDA and Ridge Regression for 2DLDA (RR-2DLDA). Regularized 2DLDA is an extension of 2DLDA with the introduction of a regularization parameter to deal with the small sample size problem. RR-2DLDA integrates ridge regression into Regularized 2DLDA to balance the distances among different classes after the transformation. These proposed algorithms overcome the limitations of 2DLDA and boost recognition accuracy. The experimental results on the Yale, PIE and FERET databases showed that RR-2DLDA is superior not only to 2DLDA but also other state-of-the-art algorithms.

A panel CUSUM test of the null of cointegration

This paper proposes a simple residual-based panel CUSUM test of the null hypothesis of cointegration. The test has a limiting normal distribution that is free of nuisance parameters, it is robust to heteroskedasticity and it allows for mixtures of cointegrated and spurious alternatives. Our Monte Carlo results suggest that the test has small-size distortions and reasonable power. In our empirical application to international R&D spillovers, we present evidence suggesting that total factor productivity is heterogeneously cointegrated with foreign and domestic R&D capital stocks. © Blackwell Publishing Ltd, 2005.

Stable feature selection with support vector machines

The support vector machine (SVM) is a popular method for classification, well known for finding the maximum-margin hyperplane. Combining SVM with l1-norm penalty further enables it to simultaneously perform feature selection and margin maximization within a single framework. However, l1-norm SVM shows instability in selecting features in presence of correlated features. We propose a new method to increase the stability of l1-norm SVM by encouraging similarities between feature weights based on feature correlations, which is captured via a feature covariance matrix. Our proposed method can capture both positive and negative correlations between features. We formulate the model as a convex optimization problem and propose a solution based on alternating minimization. Using both synthetic and real-world datasets, we show that our model achieves better stability and classification accuracy compared to several state-of-the-art regularized classification methods.

Exploiting feature relationships towards stable feature selection

Feature selection is an important step in building predictive models for most real-world problems. One of the popular methods in feature selection is Lasso. However, it shows instability in selecting features when dealing with correlated features. In this work, we propose a new method that aims to increase the stability of Lasso by encouraging similarities between features based on their relatedness, which is captured via a feature covariance matrix. Besides modeling positive feature correlations, our method can also identify negative correlations between features. We propose a convex formulation for our model along with an alternating optimization algorithm that can learn the weights of the features as well as the relationship between them. Using both synthetic and real-world data, we show that the proposed method is more stable than Lasso and many state-of-the-art shrinkage and feature selection methods. Also, its predictive performance is comparable to other methods.

A transformation-based multivariate chart to monitor process dispersion

Multivariate monitoring techniques such as multivariate control charts are used to control the processes that contain more than one correlated characteristic. Although the majority of previous researches are focused on controlling only the mean vector of multivariate processes, little work has been performed to monitor the covariance matrix. In this research, a new method is presented to detect possible shifts in the covariance matrix of multivariate processes. The basis of the proposed method is to eliminate the correlation structure between the quality characteristics by transformation technique and then use an S chart for each variable. The performance of the proposed method is then compared to the ones from other existing methods and a real case is presented.