• ETRI Journal
• /
• v.15 no.2
• /
• pp.35-51
• /
• 1993

This paper presents function approximation based on nonparametric estimation. As an estimation model of function approximation, a three layered network composed of input, hidden and output layers is considered. The input and output layers have linear activation units while the hidden layer has nonlinear activation units or kernel functions which have the characteristics of bounds and locality. Using this type of network, a many-to-one function is synthesized over the domain of the input space by a number of kernel functions. In this network, we have to estimate the necessary number of kernel functions as well as the parameters associated with kernel functions. For this purpose, a new method of parameter estimation in which linear learning rule is applied between hidden and output layers while nonlinear (piecewise-linear) learning rule is applied between input and hidden layers, is considered. The linear learning rule updates the output weights between hidden and output layers based on the Linear Minimization of Mean Square Error (LMMSE) sense in the space of kernel functions while the nonlinear learning rule updates the parameters of kernel functions based on the gradient of the actual output of network with respect to the parameters (especially, the shape) of kernel functions. This approach of parameter adaptation provides near optimal values of the parameters associated with kernel functions in the sense of minimizing mean square error. As a result, the suggested nonparametric estimation provides an efficient way of function approximation from the view point of the number of kernel functions as well as learning speed.

• The Korean Journal of Applied Statistics
• /
• v.32 no.1
• /
• pp.149-160
• /
• 2019

The electricity consumption time series data of 'A' University from July 2016 to June 2017 is analyzed via nonparametric functional data clustering since the time series data can be regarded as realization of continuous functions with dependency structure. We use a Bouveyron and Jacques (Advances in Data Analysis and Classification, 5, 4, 281-300, 2011) method based on model-based functional clustering with an FEM algorithm that assumes a Gaussian distribution on functional principal components. Clusterwise analysis is provided with cluster mean functions, densities and cluster profiles.

• Wind and Structures
• /
• v.31 no.6
• /
• pp.561-573
• /
• 2020

Unsteady self-excited forces are commonly represented by parametric models such as rational functions. However, this requires complex multiparametric nonlinear fitting, which can be a challenging task that requires know-how. This paper explores the alternative nonparametric modeling of unsteady self-excited forces based on relations between flutter derivatives. By exploiting the properties of the transfer function of linear causal systems, we show that damping and stiffness aerodynamic derivatives are related by the Hilbert transform. This property is utilized to develop exact simplified expressions, where it is only necessary to consider the frequency dependency of either the aeroelastic damping or stiffness terms but not both simultaneously. This approach is useful if the experimental data on aerodynamic derivatives that are related to the damping are deemed more accurate than the data that are related to the stiffness or vice versa. The proposed numerical models are evaluated with numerical examples and with data from wind tunnel experiments. The presented method can evaluate any continuous fitted table of interpolation functions of various types, which are independently fitted to aeroelastic damping and stiffness terms. The results demonstrate that the proposed methodology performs well. The relations between the flutter derivatives can be used to enhance the understanding of experimental modeling of aerodynamic self-excited forces for bridge decks.

• Journal of Multimedia Information System
• /
• v.4 no.4
• /
• pp.189-194
• /
• 2017

In this study, we propose a method which is based on "Image segmentation by histogram thresholding using hierarchical cluster analysis"/HCA/ and "A nonparametric approach for histogram segmentation"/NHS/. HCA method uses that all histogram bins are one cluster then it reduces cluster numbers by using distance metric. Because this method has too many clusters, it is more computation. In order to eliminate disadvantages of "HCA" method, we used "NHS" method. NHS method finds all local minima of histogram. To reduce cluster number, we use NHS method which is fast. In our approach, we combine those two methods to eliminate disadvantages of Arifin method. The proposed method is not only less computational than "HCA" method because combined method has few clusters but also it uses local minima of histogram which is computed by "NHS".

• Journal of the Korean earth science society
• /
• v.21 no.2
• /
• pp.159-167
• /
• 2000

We tested the randomness and estimated the degree of concentration for the earthquake occurrence in the Korean Peninsula by using the statistical methods for spatial data. For the randomness test, we applied both of the test statistics based method and the empirical distribution based method to the both of historical and instrumental seismicity data. It was found that the earthquake occurrences for historical and instrumental seismicity data are not random and clustered rather than scattered. A nonparametric density estimation method was used to estimate the concentration degree in the Peninsula. The earthquake occurrences show relatively high concentration on Seoul, Choongnam, Chonbook and Kyungbook areas for the historical seismicity data. Also,'L" shaped concentrations connecting Whanghaedo -the coast of Choongnam -the inland of Kyungbook area are revealed for the instrumental seismicity data.

• The Korean Journal of Applied Statistics
• /
• v.27 no.6
• /
• pp.1049-1068
• /
• 2014

Autoregressive models are used to analyze an univariate time series data; however, these methods can be inappropriate when a structural break appears in a time series since they assume that a trend is consistent. Threshold autoregressive models (popular regime-switching models) have been proposed to address this problem. Recently, the models have been extended to two regime-switching models with delay parameter. We discuss two regime-switching threshold autoregressive models from a Bayesian point of view. For a Bayesian analysis, we consider a parametric threshold autoregressive model and a nonparametric threshold autoregressive model using Dirichlet process prior. The posterior distributions are derived and the posterior inferences is performed via Markov chain Monte Carlo method and based on two Bayesian threshold autoregressive models. We present a simulation study to compare the performance of the models. We also apply models to gross domestic product data of U.S.A and South Korea.

• The Korean Journal of Applied Statistics
• /
• v.29 no.5
• /
• pp.807-822
• /
• 2016

The composite lognormal-GPD models (LN-GPD) enjoys both merits from log-normality for the body of distribution and GPD for the thick tailedness of the observation. However, in the estimation perspective, LN-GPD model performs poorly due to numerical instability. Therefore, a two-stage procedure, that estimates threshold first then estimates other parameters later, is a natural method to consider. This paper considers five nonparametric threshold estimation methods widely used in extreme value theory and compares their performance in LN-GPD parameter estimation. A simulation study reveals that simultaneous maximum likelihood estimation performs good in threshold estimation, but very poor in tail index estimation. However, the nonparametric method performs good in tail index estimation, but introduced bias in threshold estimation. Our method is illustrated to the service time of an Israel bank call center and shows that the LN-GPD model fits better than LN or GPD model alone.

• Osong Public Health and Research Perspectives
• /
• v.5 no.4
• /
• pp.187-192
• /
• 2014

Objectives: Measurement of the incidence of the human immunodeficiency virus (HIV) is very important for epidemiological studies. Here, we determined the recency period with the AxSYM avidity assay and the BED-capture enzyme immunoassay (BED-CEIA) in Korean seroconverters. Methods: Two hundred longitudinal specimens from 81 seroconverters with incident HIV infections that had been collected at the Korea National Institute of Health were subjected to the AxSYM avidity assay (cutoff = 0.8) and BED-CEIA (cutoff = 0.8). The statistical method used to estimate the recency period in recent HIV infections was nonparametric survival analyses. Sensitivity and specificity were calculated for 10-day increments from 120 days to 230 days to determine the recency period. Results: The mean recency period of the avidity assay and BED-CEIA using a survival method was 158 days [95% confidence interval (CI), 135-181 days] and 189 days (95% CI, 170-208 days), respectively. Based on the use of sensitivity and specificity, the mean recency period for the avidity assay and BED-CEIA was 150 days and 200 days, respectively. Conclusion: We determined the recency period to estimate HIV incidence in Korea. These data showed that the nonparametric survival analysis often led to shorter recency periods than analysis of sensitivity and specificity as a new method. These findings suggest that more data from seroconverters and other methodologies are needed to determine the recency period for estimating HIV incidence.

• Journal of the Korean Data and Information Science Society
• /
• v.18 no.4
• /
• pp.1005-1021
• /
• 2007

Traditional procedures for DNA microarray data analysis are to preprocess and normalize the gene expression data, and then to analyze the normalized data using statistical tests. Drawbacks of the traditional methods are: genuine biological signal may be unwillingly eliminated together with artifacts, the limited number of arrays per gene make statistical tests difficult to use the normality assumption or nonparametric method, and genes are tested independently without consideration of interrelationships among genes. A novel method using average rank in each array is proposed to eliminate such drawbacks. This average rank method monitors differentially regulated genes among genetically different groups and the selected genes are somewhat different from those selected by traditional P-value method. Addition of genes selected by the average rank method to the traditional method will provide better understanding of genetic differences of groups.

• Journal of the Korean Statistical Society
• /
• v.17 no.2
• /
• pp.61-80
• /
• 1988

For testing homogeneity of scale parameters aginst ordered alternatives, some nonparametric test statistics based on pairwise ranking method are proposed. The proposed tests are distribution-free. The asymptotic distributions of the proposed statistcs are also investigated. It is shown that the Pitman efficiencies of the proposed rank tests are the same as those of the corresponding two-sample rank tests in the scale problem. A small-sample Monte Carlo study is also performed. The results show that the proposed tests are robust and efficient.