• Journal of the Korean Data and Information Science Society
• /
• v.20 no.5
• /
• pp.949-954
• /
• 2009

The autoregressive process is applied in this paper to kernel regression in order to infer nonlinear models for predicting responses. We propose a kernel method for the autoregressive data which estimates the mean function by kernel machines. We also present the model selection method which employs the cross validation techniques for choosing the hyper-parameters which affect the performance of kernel regression. Artificial and real examples are provided to indicate the usefulness of the proposed method for the estimation of mean function in the presence of autocorrelation between data.

• MALSORI
• /
• no.66
• /
• pp.105-116
• /
• 2008

In this research, we propose a speaker identification system using a kernel method which is expected to model the non-linearity of speech features well. We have been using principal component analysis (PCA) successfully, and extended to kernel PCA, which is used for many pattern recognition tasks such as face recognition. However, we cannot use kernel PCA for speaker identification directly because the storage required for the kernel matrix grows quadratically, and the computational cost grows linearly (computing eigenvector of $l{\times}l$ matrix) with the number of training vectors I. Therefore, we use greedy kernel PCA which can approximate kernel PCA with small representation error. In the experiments, we compare the accuracy of the greedy kernel PCA with the baseline Gaussian mixture models using MFCCs and PCA. As the results with limited enrollment data show, the greedy kernel PCA outperforms conventional methods.

• 제어로봇시스템학회:학술대회논문집
• /
• 1999.10a
• /
• pp.137-140
• /
• 1999

A lot of researcher have proposed a method of kernel identifying nonlinear system by use of Wiener kernels[6-7] or Volterra kernel[5] and so on. In this research, the authors proposed a method of identifying Volterra kernels for nonlinear system by use of pseudorandom M-sequence in which a crosscorrelation function between input and output of a nonlinear system is taken[4]. we can be applied to an MISO nonlinear system or a system which depends on its input amplitude[2]. But, there exist many systems in which it is difficult to determine a Volterra kernel having the same time coordinate on the crosscorrelation function. In those cases, we have to estimate Volterra kernel by using its neighboring points[4]. In this paper, we propose a new method for not estimating but obtaining Volterra kernel having the same time coordinate using calculation between the neighboring points. Some numerical simulations show that this method is effective for obtaining higher order Volterra kernel of nonlinear control systems.

• Journal of the Korean Data and Information Science Society
• /
• v.12 no.2
• /
• pp.83-93
• /
• 2001

We investigate a kernel approach to discriminant analysis for binary classification as a machine learning point of view. Our view of the kernel approach follows support vector method which is one of the most promising techniques in the area of machine learning. As usual discriminant analysis, the kernel method can discriminate an object most likely belongs to. Moreover, it has some advantage over discriminant analysis such as data compression and computing time.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.12 no.7
• /
• pp.3251-3257
• /
• 2011

This paper presents a new Boolean extraction technique for logic synthesis. This method extracts kernel-kernel pairs as well as cokernel-kernel pairs. The given logic expressions can be translated into Boolean divisors and quotients with kernel-kernel pairs. Next, kernel intersection method provides the common sub-expressions for several logic expressions. Experimental results show the improvement in literal count over previous other extraction methods.

• Interaction and multiscale mechanics
• /
• v.2 no.3
• /
• pp.295-319
• /
• 2009

In this work, we discuss a reproducing kernel collocation method (RKCM) for solving $2^{nd}$ order PDE based on strong formulation, where the reproducing kernel shape functions with compact support are used as approximation functions. The method based on strong form collocation avoids the domain integration, and leads to well-conditioned discrete system of equations. We investigate the convergence and the computational complexity for this proposed method. An important result obtained from the analysis is that the degree of basis in the reproducing kernel approximation has to be greater than one for the method to converge. Some numerical experiments are provided to validate the error analysis. The complexity of RKCM is also analyzed, and the complexity comparison with the weak formulation using reproducing kernel approximation is presented.

• Communications for Statistical Applications and Methods
• /
• v.19 no.4
• /
• pp.517-526
• /
• 2012

In this paper a support vector method is proposed for use when the sample observations are contaminated by a normally distributed measurement error. The performance of deconvolution density estimators based on the support vector method is explored and compared with kernel density estimators by means of a simulation study. An interesting result was that for the estimation of kurtotic density, the support vector deconvolution estimator with a Gaussian kernel showed a better performance than the classical deconvolution kernel estimator.

• Communications for Statistical Applications and Methods
• /
• v.17 no.3
• /
• pp.451-457
• /
• 2010

This paper considers the problem of nonparametric deconvolution density estimation when sample observa-tions are contaminated by double exponentially distributed errors. Three different deconvolution density estima-tors are introduced: a weighted kernel density estimator, a kernel density estimator based on the support vector regression method in a RKHS, and a classical kernel density estimator. The performance of these deconvolution density estimators is compared by means of a simulation study.

• Journal of the Korean Society of Industry Convergence
• /
• v.3 no.1
• /
• pp.17-27
• /
• 2000

A factorization is an extremely important part of multi-level logic synthesis. The number of literals in a factored form is a good estimate of the complexity of a logic function. and can be translated directly into the number of transistors required for implementation. Factored forms are described as either algebraic or Boolean, according to the trade-off between run-time and optimization. A Boolean factored form contains fewer number of literals than an algebraic factored form. In this paper, we present a new method for a Boolean factorization. The key idea is to build an extended co-kernel cube matrix using co-kernel/kernel pairs and kernel/kernel pairs together. The extended co-kernel cube matrix makes it possible to yield a Boolean factored form. We also propose a heuristic method for covering of the extended co-kernel cube matrix. Experimental results on various benchmark circuits show the improvements in literal counts over the algebraic factorization based on Brayton's co-kernel cube matrix.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.22 no.5
• /
• pp.561-566
• /
• 2021

The optimum kernel size for error-distribution estimation with given error samples cannot be used in the weight adjustment of minimum Euclidean distance between error distributions (MED) algorithms. In this paper, a new adaptive kernel estimation method for convergence enhancement of MED algorithms is proposed. The proposed method uses the average rate of change in error power with respect to a small interval of the kernel width for weight adjustment of the MED learning algorithm. The proposed kernel adjustment method is applied to experiments in communication channel compensation, and performance improvement is demonstrated. Unlike the conventional method yielding a very small kernel calculated through optimum estimation of error distribution, the proposed method converges to an appropriate kernel size for weight adjustment of the MED algorithm. The experimental results confirm that the proposed kernel estimation method for MED can be considered a method that can solve the sensitivity problem from choosing an appropriate kernel size for the MED algorithm.