• International Journal of Fuzzy Logic and Intelligent Systems
• /
• v.13 no.3
• /
• pp.186-199
• /
• 2013

We pose pattern classification as a density estimation problem where we consider mixtures of generative models under partially labeled data setups. Unlike traditional approaches that estimate density everywhere in data space, we focus on the density along the decision boundary that can yield more discriminative models with superior classification performance. We extend our earlier work on the recursive estimation method for discriminative mixture models to semi-supervised learning setups where some of the data points lack class labels. Our model exploits the mixture structure in the functional gradient framework: it searches for the base mixture component model in a greedy fashion, maximizing the conditional class likelihoods for the labeled data and at the same time minimizing the uncertainty of class label prediction for unlabeled data points. The objective can be effectively imposed as individual mixture component learning on weighted data, hence our mixture learning typically becomes highly efficient for popular base generative models like Gaussians or hidden Markov models. Moreover, apart from the expectation-maximization algorithm, the proposed recursive estimation has several advantages including the lack of need for a pre-determined mixture order and robustness to the choice of initial parameters. We demonstrate the benefits of the proposed approach on a comprehensive set of evaluations consisting of diverse time-series classification problems in semi-supervised scenarios.

• The Korean Journal of Applied Statistics
• /
• v.31 no.1
• /
• pp.41-65
• /
• 2018

In Korea, as the mortality rate improves in a shorter period of time than in developed countries, it is important to consider the selection of the time series as well as the model selection in the mortality projection. Therefore, this study proposed a method using the multiple regression model in respect to the selection of the time series period. In addition, we investigate the problems that arise when various time series are used based on the Lee-Carter (LC) model, the kinds of LC model along with Lee-Miller (LM) and Booth-Maindonald-Smith (BMS), and the non-parametric model such as functional data model (FDM) and Coherent FDM, and examine differences in the age-specific mortality rate and life expectancy projection. Based on the analysis results, the age-specific mortality rate and predicted life expectancy of men and women are calculated for the year 2030 for each model. We also compare the mortality rate and life expectancy of the next generation provided by Korean Statistical Information Service (KOSIS).

• 한국데이터정보과학회:학술대회논문집
• /
• 2005.10a
• /
• pp.135-138
• /
• 2005

In this paper, we consider the problem of testing for parameter change in time series models based on a cusum of squares. Although the test procedure is well-established for the mean and variance in time series models, a general parameter case was not discussed in literatures. Therefore, here we develop the cusum of squares type test for parameter change in a more general framework. As an example, we consider the change of the parameters in an RCA(1) model. Simulation results are reported for illustration.

• Proceedings of the Korean Statistical Society Conference
• /
• 2001.11a
• /
• pp.185-189
• /
• 2001

In this paper, we consider the problem of testing for parameter changes in time series models based on a sequential test. Although the test procedure is well-established for the mean and variance change, a general parameter case has not been discussed in the literature. Therefore, we develop a sequential test for parameter changes in a more general framework.

• Proceedings of the Korean Vacuum Society Conference
• /
• 2015.08a
• /
• pp.222.1-222.1
• /
• 2015

Recently, graphene quantum dots (GQDs) have attracted great attention due to various properties including cost-effectiveness of synthesis, low toxicity, and high photostability. Nevertheless, the origins of photoluminescence (PL) from GQDs are unclear because of extrinsic states of the impurities, disorder structures, and oxygen-functional groups. Therefore, to utilize GQDs in various applications, their optical properties generated from the extrinsic states should be understood. In this work, we have focused on the effect of oxygen-functional groups in PL of the GQDs. The GQDs with nanoscale and single layer are synthesized by employing graphite nanoparticles (GNPs) with 4 nm. The series of GQDs with different amount of oxygen-functional groups were prepared by the chain of chemical oxidation and reduction process. The fabrication of a series of graphene oxide QDs (GOQDs) with different amounts of oxygen-contents is first reported by a direct oxidation route of GNPs. In addition, for preparing a series of reduced GOQDs (rGOQDs), we employed the conventional chemical reduction to GOQDs solution and controlled the amount of reduction agents. The GOQDs and rGOQDs showed irreversible PL properties even though both routes have similar amount of oxyen-functional groups. In the case of a series of GOQDs, the PL spectrum was clearly redshifted into blue and green-yellowish color. On the other hand, the PL spectrum of rGOQDs did not change significantly. By various optical measurement such as the PL excitation, UV-vis absorbance, and time-resolved PL, we could verify that their PL mechanisms of GOQDs and rGOQDs are closely associated with different atomic structures formed by chemical oxidation and reduction. Our study provides an important insights for understanding the optical properties of GQDs affected by oxygen-functional groups. [1]

• Journal of the Korean Society of Costume
• /
• v.53 no.8
• /
• pp.53-63
• /
• 2003

Under the social conditions that an understanding of a men change socially and men's interests in their appearances are on the increase, this research intends to study and analyze the changes of people's sensibility of values on ads for men's cosmetics, which is said to be performed the social, cultural function reflecting the social consciousness, the value and the idea creating a new one. The documents for the research was sindonga. The documents were analyzed quantitatively in regard of contents of a linguistic expression and a visual expression showing on advertisements for man's cosmetics, a time series analysis. As a result of analyzing the consumption value through headline of an ad, the emotional value took the highest percentage extending whole periods. Beside emotional value, a time series analysis showed that social values appeared to the major themes in the 1970s, and functional values was important themes in 1980s, and the social values and functional values appeared equally during 1990-2002.

• Communications for Statistical Applications and Methods
• /
• v.31 no.1
• /
• pp.87-104
• /
• 2024

Most functional magnetic resonance imaging (fMRI) studies in resting state have assumed that the functional connectivity (FC) between time series from distinct brain regions is constant. However, increased interest has recently been in quantifying possible dynamic changes in FC during fMRI experiments. FC study may provide insight into the fundamental workings of brain networks to brain activity. In this work, we focus on the specific problem of estimating the dynamic behavior of pairwise correlations between time courses extracted from two different brain regions. We compare the sliding-window techniques such as moving average (MA) and exponentially weighted moving average (EWMA), dynamic causality with vector autoregressive (VAR) model, dynamic conditional correlation (DCC) based on volatility, and the proposed alternative methods to use differencing and recursive residuals. We investigate the properties of those techniques in a series of simulation studies. We also provide an application with major depressive disorder (MDD) patient fMRI data to demonstrate studying dynamic correlations.

• The Korean Journal of Applied Statistics
• /
• v.33 no.3
• /
• pp.297-308
• /
• 2020

We focus on the functional autoregressive conditional heteroscedasticity (fARCH) modelling to analyze intraday volatilities based on high frequency financial time series. Multivariate volatility models are investigated to approximate fARCH(1). A formula of multi-step ahead volatilities for fARCH(1) model is derived. As an application, in implementing fARCH(1), a choice of appropriate time interval for the intraday return is discussed. High frequency KOSPI data analysis is conducted to illustrate the main contributions of the article.

• Journal of the Korean Data and Information Science Society
• /
• v.26 no.4
• /
• pp.885-894
• /
• 2015

It is necessary to forecast the electricity demand for reliable and effective operation of the power system. In this study, we try to categorize a functional data, the mean curve in accordance with the time of daily power demand pattern. The data were collected between January 1, 2009 and December 31, 2011. And it were converted to time series data consisting of seasonal components and error component through log transformation and removing trend. Functional clustering by Ma et al. (2006) are applied and parameters are estimated using EM algorithm and generalized cross validation. The number of clusters is determined by classifying holidays or weekdays. Monday, weekday (Tuesday to Friday), Saturday, Sunday or holiday and season are described the mean curve of daily power demand pattern.

• The Korean Journal of Applied Statistics
• /
• v.32 no.1
• /
• pp.69-81
• /
• 2019

We consider bivariate long range dependent (LRD) time series forecasting using a deep learning method. A long short-term memory (LSTM) network well-suited to time series data is applied to forecast bivariate time series; in addition, we compare the forecasting performance with bivariate fractional autoregressive integrated moving average (FARIMA) models. Out-of-sample forecasting errors are compared with various performance measures for functional MRI (fMRI) data and daily realized volatility data. The results show a subtle difference in the predicted values of the FIVARMA model and VARFIMA model. LSTM is computationally demanding due to hyper-parameter selection, but is more stable and the forecasting performance is competitively good to that of parametric long range dependent time series models.