Local Kernel Research Articles

Twin Support Vector Machines Based on the Mixed Kernel function

The efficiency and performance of the Twin Support Vector Machines (TWSVM) are better than the traditional support vector machines when it deals with the problems. However, it also has the problem of selecting kernel functions. Generally, TWSVM selects the Gaussian radial basis kernel function. Although it has a strong learning ability, its generalization ability is relatively weak. In a certain extent, this will limit the performance of TWSVM. In order to solve the problem of selecting kernel functions in TWSVM, we propose the twin support vector machines based on the mixed kernel function (MK-TWSVM) in this paper. To make full use of the learning ability of local kernel functions and the excellent generalization ability of global kernel functions, MK-TWSVM selects a global kernel function and a local kernel function to construct a mixed kernel function which has the better performance. The experimental results indicate that the mixed kernel function makes TWSVM have the good learning ability and generalization ability. So it improves the performance of TWSVM.

Mixed Kernel Twin Support Vector Machines Based on the Shuffled Frog Leaping Algorithm

The efficiency and performance of Twin Support Vector Machines (TWSVM) is better than the traditional support vector machines when it deals with the problems. However, it also has some problems. As the same as the traditional support vector machines, its parameters are difficult to be appointed and it is not easy to select the appropriate kernel function. TWSVM generally selects the Gaussian radial basis kernel function. Although its learning ability is very strong, its generalization ability is relatively weak. To a certain extent, this limits the performance of TWSVM .In order to solve these two problems, in this paper, we propose the Mixed Kernel Twin Support Vector Machines based on the shuffled frog leaping algorithm (SFLA-MK-TWSVM). To make full use of both the excellent generalization ability of global kernel function s and the learning ability of local kernel function s , SFLA-MK-TWSVM constructs a mixed kernel which has better performance. Then SFLA-MK-TWSVM uses the shuffled frog leaping algorithm to determine the parameters of both TWSVM and the mixed kernel function to further improve the performance of TWSVM. The experimental results indicate that SFLA-MK-TWSVM significantly improves the classification accuracy of TWSVM.

Protein Remote Homology Detection by Combining Profile-based Protein Representation with Local Alignment Kernel

Protein remote homology detection has attracted a great deal of interest as it is one of the most important problems in bioinformatics. Profile-based methods recently achieve the state-of-the-art performance. A key step to improve the performance of these methods is to find a suitable approach to use the evolutionary information in the profiles. In this study, we propose the profile-based protein representation to extract the evolutionary information from frequency profiles. In this approach, the frequency profiles calculated from the multiple sequence alignments outputted by PSI-BLAST are converted into several profile-based proteins and then the local alignment kernel (LA) is combined with these profile-based proteins for the prediction. Our experiments on a well-known benchmark show that the proposed approach can significantly improve the predictive performance.

Top-Down Visual Saliency Detection in Optical Satellite Images Based on Local Adaptive Regression Kernel

This paper proposes a novel top-down visual saliency detection method for optical satellite images using local adaptive regression kernels. This method provides a saliency map by measuring the likeness of image patches to a given single template image. The local adaptive regression kernel (LARK) is used as a descriptor to extract feature and compare against analogous feature from the target image. A multi-scale pyramid of the target image is constructed to cope with large-scale variations. In addition, accounting for rotation variations, the histogram of kernel orientation is employed to estimate the rotation angle of image patch, and then comparison is performed after rotating the patch by the estimated angle. Moreover, we use the bounded partial correlation (BPC) to compare features between image patches and the template so as to rapidly generate the saliency map. Experiments were performed in optical satellite images to find airplanes, and experimental results demonstrate that the proposed method is effective and robust in complex scenes.

Video Object Tracking Mechanism

The video object mechanism is based on the appearance representation of the object. Tracking objects in a video has applications in video surveillance and other some applications. In previous method does not handle the case of full occlusion. The case of full occlusion can be handled by setting an LSK similarity threshold, which stops tracking when the object is lost. The objects can be tracked based on the compression domain and pixel domain. The spatial information of the object that is being tracked is retrieved using Local Steering Kernels. The Color features are extracted using color histograms. The features similarities between a candidate object ROI and the object ROI in the previous frame and the last stored object instance in the object model are measured. For extracting the foreground extended Kalman filter is used. This increases the tracking efficiency of the algorithm. Index Terms: Color histograms, localsteeringkernels, and visualobjecttracking.

SNSequate: Standard and Nonstandard Statistical Models and Methods for Test Equating

Equating is a family of statistical models and methods that are used to adjust scores on two or more versions of a test, so that the scores from different tests may be used interchangeably. In this paper we present the R package SNSequate which implements both standard and nonstandard statistical models and methods for test equating. The package construction was motivated by the need of having a modular, simple, yet comprehensive, and general software that carries out traditional and new equating methods. SNSequate currently implements the traditional mean, linear and equipercentile equating methods, as well as the mean-mean, mean-sigma, Haebara and Stocking-Lord item response theory linking methods. It also supports the newest methods such as local equating, kernel equating, and item response theory parameter linking methods based on asymmetric item characteristic functions. Practical examples are given to illustrate the capabilities of the software. A list of other programs for equating is presented, highlighting the main differences between them. Future directions for the package are also discussed.

Outlier Detection Based on Local Kernel Regression for Instance Selection

In this paper, we propose an outlier detection approach based on local kernel regression for instance selection. It evaluates the reconstruction error of instances by their neighbors to identify the outliers. Experiments are performed on the synthetic and real data sets to show the efficacy of the proposed approach in comparison with the existing counterparts.

SigHunt: horizontal gene transfer finder optimized for eukaryotic genomes

Genomic islands (GIs) are DNA fragments incorporated into a genome through horizontal gene transfer (also called lateral gene transfer), often with functions novel for a given organism. While methods for their detection are well researched in prokaryotes, the complexity of eukaryotic genomes makes direct utilization of these methods unreliable, and so labour-intensive phylogenetic searches are used instead. We present a surrogate method that investigates nucleotide base composition of the DNA sequence in a eukaryotic genome and identifies putative GIs. We calculate a genomic signature as a vector of tetranucleotide (4-mer) frequencies using a sliding window approach. Extending the neighbourhood of the sliding window, we establish a local kernel density estimate of the 4-mer frequency. We score the number of 4-mer frequencies in the sliding window that deviate from the credibility interval of their local genomic density using a newly developed discrete interval accumulative score (DIAS). To further improve the effectiveness of DIAS, we select informative 4-mers in a range of organisms using the tetranucleotide quality score developed herein. We show that the SigHunt method is computationally efficient and able to detect GIs in eukaryotic genomes that represent non-ameliorated integration. Thus, it is suited to scanning for change in organisms with different DNA composition. Source code and scripts freely available for download at http://www.iba.muni.cz/index-en.php?pg=research-data-analysis-tools-sighunt are implemented in C and R and are platform-independent. 376090@mail.muni.cz or martinkova@ivb.cz.

Estimation of environmental optima and tolerances of diatoms using multifactor multiplicative modeling

We demonstrated the use of a multifactor multiplicative regression model to study the ecology of diatoms. A dataset of 174 samples of diatoms matched with environmental variables, which were collected from similar typology streams of three major watersheds in Ethiopia was used to build the model. Tropical rivers experience wet and dry season dynamics with highest levels of pollution at low water levels. In extreme polluted sites, diatoms are still present whereas other bio-indicator groups such as fishes, macrophytes and even macro-invertebrates are absent. The distribution of selected model diatom taxa (widely known pollution tolerant and sensitive taxa) in relation to environmental descriptors was studied by fitting non-parametric multiplicative regression (NPMR) to species relative abundances. A local mean estimator and Gaussian kernel functions were used to construct the models. The models reported here are those showing the best fit for particular number of predictors. Predictors were added until the cross-validated coefficient of multiple determination (×R2) increased at least by 5% so that the model with the highest number of predictors had the highest ×R2. We used bootstrap sampling to validate the model. NPMR a powerful model to study autoecology allows us to question the assumption that a suitable indicator species exhibits approximately bell-shaped curve with a single optimum for the probability density function of a species along environmental gradients. Only a few number of taxa showed such an optima whereas multi-interaction of diatom species could cause linear, skewed or non-unimodal, bimodal and multimodal responses. Therefore, modeling of species response to environmental gradients can be best explored without predefining the response curve and considering simultaneous multiple factor interaction in a multiplicative way.

Hedonic Estimation with Unobserved Characteristics: An Application to the Housing Market in Santiago, Chile

We estimate a hedonic price function for the characteristics of housing units in Santiago, Chile, with the aim to capture unobserved utility-generating attributes for each unit. We used for this purpose two different semiparametric regression methods: a local linear kernel regression with a single unobservable index variable, similar to what has been implemented by Bajari and Kahn (2005), and a partially linear in the parameters, partially nonparametric method as first proposed by Robinson (1988). The data used comes from a cross-sectional sample of housing units in Santiago, Chile in 2006. Interestingly, the results obtained for the unobservable components are similar to each other and to ordinary least squares.

An adaptive bandwidth nonlocal means image denoising in wavelet domain

This paper proposes a new wavelet domain denoising algorithm. In the results of conventional wavelet domain denoising methods, ringing artifacts or wavelet-shaped noises are sometimes observed due to thresholding of small but important coefficients or due to generation of large coefficients in flat areas. In this paper, nonlocal means filtering is applied to each subband of wavelet decomposition, which can keep small coefficients and does not generate unwanted large coefficients. Since the performance of nonlocal means filtering depends on the appropriate kernel bandwidth, we also propose a method to find global and local kernel bandwidth for each subband. In comparison with conventional methods, the proposed method shows lower PSNR than BM3D when pseudo white Gaussian noise is added, but higher PSNR than the spatial nonlocal means filtering and wavelet thresholding methods. For the mixture noise or Poisson noise, which may better explain the real noise from camera sensors, the proposed method shows better or comparable results than the state-of-the-art methods. Also, it is believed that the proposed method shows better subjective quality for the noisy images captured in the low-illumination conditions.

Object Recognition Based on Local Steering Kernel and SVM

The proposed method is to recognize objects based on application of Local Steering Kernels (LSK) as Descriptors to the image patches. In order to represent the local properties of the images, patch is to be extracted where the variations occur in an image. To find the interest point, Wavelet based Salient Point detector is used. Then, Local Steering Kernel is applied to the resultant pixels in order to obtain the most promising features. The features extracted will be over complete; so, in order to reduce dimensionality, Principal Component Analysis (PCA) is applied. Further, the sparse histogram is taken over the PCA output. The classifier used here is Support Vector Machine (SVM) Classifier. Bench mark database used is UIUC car database and the results obtained are satisfactory. The results obtained using LSK kernel is compared by varying parameters such as patch size, number of salient points/patches, smoothing parameter and scaling parameter.

Saliency detection based on integrated features

This paper presents a novel computational model for saliency detection. The proposed model utilizes feature level fusion method to integrate different kinds of visual features. The integrated features are used to measure saliency, so no separate feature conspicuity maps, or the subsequent combination of them is needed in our model. Then, the new model combines the local and global measurements for estimating saliency (termed LGMES) by using local and global kernel density estimations during the saliency computation process. Experimental results on two human eye fixation datasets demonstrate that the proposed model outperforms the state-of-the-art methods. Meanwhile, the proposed saliency measurement is more efficient than those methods using separately local or global measurements.

A second-order numerical method for elliptic equations with singular sources using local filter

The presence of Dirac delta function in differential equation can lead to a discontinuity, which may degrade the accuracy of related numerical methods. To improve the accuracy, a second-order numerical method for elliptic equations with singular sources is introduced by employing a local kernel filter. In this method, the discontinuous equation is convoluted with the kernel function to obtain a more regular one. Then the original equation is replaced by this filtered equation around the singular points, to obtain discrete numerical form. The unchanged equations at the other points are discretized by using a central difference scheme. 1D and 2D examples are carried out to validate the correctness and accuracy of the present method. The results show that a second-order of accuracy can be obtained in the filtering framework with an appropriate integration rule. Furthermore, the present method does not need any jump condition, and also has extremely simple form that can be easily extended to high dimensional cases and complex geometry.

On the Short-Range Behavior of Correlated Pair Functions from the Adiabatic-Connection Fluctuation-Dissipation Theorem of Density-Functional Theory.

The short-range behavior of correlated pair functions from the adiabatic-connection fluctuation-dissipation theorem (ACFD) of density functional theory (DFT) employing local exchange-correlation kernels has been analyzed. It has been found that for large basis sets the pair function exhibits unphysical humps for small interelectronic distances if the adiabatic local density approximation kernel is used in the ACFD scheme (this method is termed ACFD/ALDA in this work). However, up to basis set sizes of quadruple-ζ type quality, the correlated pair function of ACFD/ALDA behaves physically correct and the method yields reasonable results for atomization energies, ionization potentials, and intermolecular interaction energies. In order to correct the deficiencies of the pair function of ACFD/ALDA for large basis sets, a short-range correction scheme has been devised on the basis of a combination of the ACFD/ALDA pair function for the large distance regime with a proper physically correctly behaving pair function at smaller distances. While this approach, termed as ACFD/ALDAcorr, practically yields results close to those of the ACFD/ALDA method for finite basis sets, it enables basis set extrapolation techniques and thus can take dynamic correlation effects fully into account in contrast to the ACFD/ALDA approach. This work also presents an efficient density-fitting algorithm to compute the ACFD correlation energies that enables the calculation of correlation energies of extended molecular systems.

Nonparametric modeling and analysis of association between Huntington's disease onset and CAG repeats

Huntington's disease (HD) is a neurodegenerative disorder with a dominant genetic mode of inheritance caused by an expansion of CAG repeats on chromosome 4. Typically, a longer sequence of CAG repeat length is associated with increased risk of experiencing earlier onset of HD. Previous studies of the association between HD onset age and CAG length have favored a logistic model, where the CAG repeat length enters the mean and variance components of the logistic model in a complex exponential-linear form. To relax the parametric assumption of the exponential-linear association to the true HD onset distribution, we propose to leave both mean and variance functions of the CAG repeat length unspecified and perform semiparametric estimation in this context through a local kernel and backfitting procedure. Motivated by including family history of HD information available in the family members of participants in the Cooperative Huntington's Observational Research Trial (COHORT), we develop the methodology in the context of mixture data, where some subjects have a positive probability of being risk free. We also allow censoring on the age at onset of disease and accommodate covariates other than the CAG length. We study the theoretical properties of the proposed estimator and derive its asymptotic distribution. Finally, we apply the proposed methods to the COHORT data to estimate the HD onset distribution using a group of study participants and the disease family history information available on their family members.

Predicting anticancer peptides with Chou′s pseudo amino acid composition and investigating their mutagenicity via Ames test

Cancer is an important reason of death worldwide. Traditional cytotoxic therapies, such as radiation and chemotherapy, are expensive and cause severe side effects. Currently, design of anticancer peptides is a more effective way for cancer treatment. So there is a need to develop a computational method for predicting the anticancer peptides. In the present study, two methods have been developed to predict these peptides using support vector machine (SVM) as a powerful machine learning algorithm. Classifiers have been applied based on the concept of Chou′s pseudo-amino acid composition (PseAAC) and local alignment kernel. Since a number of HIV-1 proteins have cytotoxic effect, therefore we predicted the anticancer effect of HIV-1 p24 protein with these methods. After the prediction, mutagenicity of 2 anticancer peptides and 2 non-anticancer peptides was investigated by Ames test. Our results show that, the accuracy and the specificity of local alignment kernel based method are 89.7% and 92.68%, respectively. The accuracy and specificity of PseAAC-based method are 83.82% and 85.36%, respectively. By computational analysis, out of 22 peptides of p24 protein, 4 peptides are anticancer and 18 are non-anticancer. In the Ames test results, it is clear that anticancer peptides (ARP788.8 and ARP788.21) are not mutagenic. Therefore the results demonstrate that the described computation methods are useful to identify potential anticancer peptides, which are worthy of further experimental validation and 2 peptides (ARP788.8 and ARP788.21) of HIV-1 p24 protein can be used as new anticancer candidates without mutagenicity.

Robust variable selection for the varying coefficient model based on composite L 1–L 2 regression

The varying coefficient model (VCM) is an important generalization of the linear regression model and many existing estimation procedures for VCM were built on L 2 loss, which is popular for its mathematical beauty but is not robust to non-normal errors and outliers. In this paper, we address the problem of both robustness and efficiency of estimation and variable selection procedure based on the convex combined loss of L 1 and L 2 instead of only quadratic loss for VCM. By using local linear modeling method, the asymptotic normality of estimation is driven and a useful selection method is proposed for the weight of composite L 1 and L 2. Then the variable selection procedure is given by combining local kernel smoothing with adaptive group LASSO. With appropriate selection of tuning parameters by Bayesian information criterion (BIC) the theoretical properties of the new procedure, including consistency in variable selection and the oracle property in estimation, are established. The finite sample performance of the new method is investigated through simulation studies and the analysis of body fat data. Numerical studies show that the new method is better than or at least as well as the least square-based method in terms of both robustness and efficiency for variable selection.

Design of k‐space channel combination kernels and integration with parallel imaging

In this work, a new method is described for producing local k-space channel combination kernels using a small amount of low-resolution multichannel calibration data. Additionally, this work describes how these channel combination kernels can be combined with local k-space unaliasing kernels produced by the calibration phase of parallel imaging methods such as GRAPPA, PARS and ARC. Experiments were conducted to evaluate both the image quality and computational efficiency of the proposed method compared to a channel-by-channel parallel imaging approach with image-space sum-of-squares channel combination. Results indicate comparable image quality overall, with some very minor differences seen in reduced field-of-view imaging. It was demonstrated that this method enables a speed up in computation time on the order of 3-16X for 32-channel data sets. The proposed method enables high quality channel combination to occur earlier in the reconstruction pipeline, reducing computational and memory requirements for image reconstruction.

Local polynomial quasi-likelihood regression with truncated and dependent data

The local polynomial quasi-likelihood estimation has several good statistical properties such as high minimax efficiency and adaptation of edge effects. In this paper, we construct a local quasi-likelihood regression estimator for a left truncated model, and establish the asymptotic normality of the proposed estimator when the observations form a stationary and α-mixing sequence, such that the corresponding result of Fan et al. [Local polynomial kernel regression for generalized linear models and quasilikelihood functions, J. Amer. Statist. Assoc. 90 (1995), pp. 141–150] is extended from the independent and complete data to the dependent and truncated one. Finite sample behaviour of the estimator is investigated via simulations too.