• Journal of the Korea Convergence Society
• /
• v.9 no.4
• /
• pp.47-55
• /
• 2018

In this paper, we propose a novel face recognition(FR) method that takes advantage of combining weighted deep local features extracted from multiple Deep Convolutional Neural Networks(DCNNs) learned with a set of facial local regions. In the proposed method, the so-called weighed deep local features are generated from multiple DCNNs each trained with a particular face local region and the corresponding weight represents the importance of local region in terms of improving FR performance. Our weighted deep local features are applied to Joint Bayesian metric learning in conjunction with Nearest Neighbor(NN) Classifier for the purpose of FR. Systematic and comparative experiments show that our proposed method is robust to variations in pose, illumination, and expression. Also, experimental results demonstrate that our method is feasible for improving face recognition performance.

• Journal of KIISE:Software and Applications
• /
• v.31 no.12
• /
• pp.1569-1580
• /
• 2004

One commonly used approach to deal with uncertainty is Bayesian network which represents joint probability distributions of domain. There are some attempts to team the structure of Bayesian networks automatically and recently many researchers design structures of Bayesian network using evolutionary algorithm. However, most of them use the only one fittest solution in the last generation. Because it is difficult to combine all the important factors into a single evaluation function, the best solution is often biased and less adaptive. In this paper, we present a method of generating diverse Bayesian network structures through fitness sharing and combining them by Bayesian method for adaptive inference. In order to evaluate performance, we conduct experiments on learning Bayesian networks with artificially generated data from ASIA and ALARM networks. According to the experiments with diverse conditions, the proposed method provides with better robustness and adaptation for handling uncertainty.

• Journal of the Korean Operations Research and Management Science Society
• /
• v.39 no.4
• /
• pp.51-69
• /
• 2014

Times to multiple events (TMEs) are a major data type in large-scale business and medical data. Despite its importance, the analysis of TME data has not been well studied because of the analysis difficulty from censoring of observation. To address this difficulty, we have developed a Bayesian-based multivariate survival analysis method, which can successfully estimate the joint probability density of survival times. In this work, we extended this method for the analysis of precedence, dependency and causality among multiple events. We applied this method to the electronic health records of 2,111 patients in a children's hospital in the US and the proposed analysis successfully shows the relation between times to two types of hospital visits for different medical issues. The overall result implies the usefulness of the multivariate survival analysis method in large-scale big data in a variety of areas including marketing, human resources, and e-commerce. Lastly, we suggest our future research directions based multivariate survival analysis method.

• Communications for Statistical Applications and Methods
• /
• v.24 no.5
• /
• pp.507-518
• /
• 2017

Volatility plays a crucial role in theory and applications of asset pricing, optimal portfolio allocation, and risk management. This paper proposes a combined model of autoregressive moving average (ARFIMA), generalized autoregressive conditional heteroscedasticity (GRACH), and skewed-t error distribution to accommodate important features of volatility data; long memory, heteroscedasticity, and asymmetric error distribution. A fully Bayesian approach is proposed to estimate the parameters of the model simultaneously, which yields parameter estimates satisfying necessary constraints in the model. The approach can be easily implemented using a free and user-friendly software JAGS to generate Markov chain Monte Carlo samples from the joint posterior distribution of the parameters. The method is illustrated by using a daily volatility index from Chicago Board Options Exchange (CBOE). JAGS codes for model specification is provided in the Appendix.

• Nuclear Engineering and Technology
• /
• v.53 no.2
• /
• pp.357-367
• /
• 2021

In general, common cause failures (CCFs) have been modeled with the assumption that components within the same group are symmetric. This assumption reduces the number of parameters required for the CCF probability estimation and allows us to use a parametric model, such as the alpha factor model. Although there are various asymmetric conditions in nuclear power plants (NPPs) to be addressed, the traditional CCF models are limited to symmetric conditions. Therefore, this paper proposes the copulabased CCF model to deal with asymmetric as well as symmetric CCFs. Once a joint distribution between the components is constructed using copulas, the proposed model is able to provide the probability of common cause basic events (CCBEs) by formulating a system of equations without symmetry assumptions. In addition, Bayesian inferences for the parameters of the marginal and copula distributions are introduced and Markov Chain Monte Carlo (MCMC) algorithms are employed to sample from the posterior distribution. Three example cases using simulated data, including asymmetry conditions in total failure probabilities and/or dependencies, are illustrated. Consequently, the copula-based CCF model provides appropriate estimates of CCFs for asymmetric conditions. This paper also discusses the limitations and notes on the proposed method.

• The Korean Journal of Applied Statistics
• /
• v.33 no.2
• /
• pp.123-136
• /
• 2020

It is common to encounter correlated multiple outcomes measured on the same subject in various research fields. In developmental toxicity studies, presence of malformed pups and fetal weight are measured on the pregnant dams exposed to different levels of a toxic substance. Joint analysis of such two outcomes can result in more efficient inferences than separate models for each outcome. Most methods for joint modeling assume a normal distribution as random effects. However, in developmental toxicity studies, the response distributions may change irregularly in location and shape as the level of toxic substance changes, which may not be captured by a normal random effects model. Motivated by applications in developmental toxicity studies, we propose a Bayesian joint model for binary and continuous outcomes. In our model, we incorporate a skewed logit model for the binary outcome to allow the response distributions to have flexibly in both symmetric and asymmetric shapes on the toxic levels. We apply our proposed method to data from a developmental toxicity study of diethylhexyl phthalate.

• Transactions of the Korean Society of Mechanical Engineers A
• /
• v.41 no.11
• /
• pp.1055-1064
• /
• 2017

Since a device such as a rocket motor requires very high reliability, a reasonable reliability design process is essential. However, Korea has implemented a design method for applying a safety factor to each component. In classic reliability analysis, input variables such as mean and standard deviation, used in the limit state function, are treated as deterministic values. Because the mean and standard deviation are determined by a small amount of data, this approach could lead to inaccurate results. In this study, reliability analysis is performed for bolted joints and o-ring seals, and the Bayesian approach is used to statistically estimate the input variables. The estimated variables and failure probability, calculated by the reliability analysis, are derived in the form of probability distributions.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.1 no.2
• /
• pp.16-34
• /
• 1991

It has been widely accepted that expert systems must reason from multiple sources of information that is to some degree evidential - uncertain, imprecise, and occasionally inaccurate - called evidential information. Evidence theory (Dempster/Shafet theory) provides one of the most general framework for representing evidential information compared to its alternatives such as Bayesian theory or fuzzy set theory. Many expert system applications require evidence to be specified in the continuous domain - such as time, distance, or sensor measurements. However, the existing evidence theory does not provide an effective approach for dealing with evidence about continuous variables. As an extension to Strat's pioneeiring work, this paper provides a new combination rule, a new method for mass function transffrmation, and a new method for rendering joint mass fuctions which are of great utility in evidence theory in the continuous domain.

• Journal of the Korean Data and Information Science Society
• /
• v.28 no.1
• /
• pp.207-215
• /
• 2017

In this paper we study pooling effects in Bayesian testing procedures of independence for contingency tables from small areas. In small area estimation setup, we typically use a hierarchical Bayesian model for borrowing strength across small areas. This techniques of borrowing strength in small area estimation is used to construct a Bayes test of independence for contingency tables from small areas. In specific, we consider the methods of direct or indirect pooling in multinomial models through Dirichlet priors. We use the Bayes factor (or equivalently the ratio of the marginal likelihoods) to construct the Bayes test, and the marginal density is obtained by integrating the joint density function over all parameters. The Bayes test is computed by performing a Monte Carlo integration based on the method proposed by Nandram and Kim (2002).

• Korean Management Science Review
• /
• v.32 no.2
• /
• pp.69-78
• /
• 2015

This paper considers a Bayesian Poisson model for multivariate count data using multiplicative rates. More specifically we compose the parameter for overall arrival rates by the product of two parameters, a common effect and an individual effect. The common effect is composed of autoregressive evolution of the parameter, which allows for analysis on seasonal effects on all multivariate time series. In addition, analysis on individual effects allows the researcher to differentiate the time series by whatevercharacterization of their choice. This type of model allows the researcher to specifically analyze two different forms of effects separately and produce a more robust result. We illustrate a simple MCMC generation combined with a Gibbs sampler step in estimating the posterior joint distribution of all parameters in the model. On the whole, the model presented in this study is an intuitive model which may handle complicated problems, and we highlight the properties and possible applications of the model with an example, analyzing real time series data involving customer arrivals to a large retail store.