• Korea journal of population studies
• /
• v.21 no.1
• /
• pp.162-183
• /
• 1998

This study examines the sampling bias that may have resulted from the large number of missing observations. Despite well-designed and reliable sampling procedures, the observed sample values in DSFH(Demographic Survey on Changes in Family and Household Structure, Japan) included many missing observations. The head administerd survey method of DSFH resulted in a large number of missing observations regarding characteristics of elderly non-head parents and their children. In addition, the response probability of a particular item in DSFH significantly differs by characteristics of elderly parents and their children. Furthermore, missing observations of many items occurred simultaneously. This complex pattern of missing observations critically limits the ability to produce an unbiased analysis. First, the large number of missing observations is likely to cause a misleading estimate of the standard error. Even worse, the possible dependency of missing observations on their latent values is likely to produce biased estimates of covariates. Two models are employed to solve the possible inference biases. First, EM algorithm is used to infer the missing values based on the knowledge of the association between the observed values and other covariates. Second, a selection model was employed given the suspicion that the probability of missing observations of proximity depends on its unobserved outcome.

• Journal of Korea Water Resources Association
• /
• v.46 no.8
• /
• pp.857-871
• /
• 2013

The interest in rainfall observation and forecasting using remote sensing method like RADAR (Radio Detection and Ranging) and satellite image is increased according to increased damage by rapid weather change like regional torrential rain and flash flood. In this study, the basin runoff was calculated using adaptive neuro-fuzzy technique, one of the data driven model and MAPLE (McGill Algorithm for Precipitation Nowcasting by Lagrangian Extrapolation) forecasted precipitation data as one of the input variables. The flood estimation method using neuro-fuzzy technique and RADAR forecasted precipitation data was evaluated. Six rainfall events occurred at flood season in 2010 and 2011 in Chungju Reservoir basin were used for the input data. The flood estimation results according to the rainfall data used as training, checking and testing data in the model setup process were compared. The 15 models were composed of combination of the input variables and the results according to change of clustering methods were compared and analysed. From this study was that using the relatively larger clustering radius and the biggest flood ever happened for training data showed the better flood estimation. The model using MAPLE forecasted precipitation data showed relatively better result at inflow estimation Chungju Reservoir.

• Journal of the Earthquake Engineering Society of Korea
• /
• v.9 no.2 s.42
• /
• pp.37-46
• /
• 2005

To date, many viable smart base isolation systems have been proposed and investigated. In this study, a novel friction pendulum system (FPS) and an MR damper are employed as the isolator and supplemental damping device, respectively, of the smart base isolation system. A fuzzy logic controller (FLC) is used to modulate the MR damper because the FLC has an inherent robustness and ability to handle non linearities and uncertainties. A genetic algorithm (GA) is used for optimization of the FLC. The main purpose of employing a GA is to determine appropriate fuzzy control rules as well to adjust parameters of the membership functions. To this end, a GA with a local improvement mechanism is applied. This method is efficient in improving local portions of chromosomes. Neuro fuzzy models are used to represent dynamic behavior of the MR damper and FPS. Effectiveness of the proposed method for optimal design of the FLC is judged based on computed responses to several historical earthquakes. It has been shown that the proposed method can find optimal fuzzy rules and the GA optimized FLC outperforms not only a passive control strategy but also a human designed FLC and a conventional semi active control algorithm.

• KSCE Journal of Civil and Environmental Engineering Research
• /
• v.33 no.2
• /
• pp.781-790
• /
• 2013

This research focuses on development of the conceptual cost estimation models for I.L.M box girder bridge. The current conceptual cost estimation for public construction projects is dependent on governmental average unit price references which has been regarded as inaccurate and unreliable by many experts. Therefore, there have been strong demands for developing a better way of conceptual cost estimating methods. This research has proposed three different conceptual cost estimating method for a P.S.C. girder bridge built with the I.L.M method. Model (I) attempts to seek the proper breakdown of standard works that are accountable for more than 95 percentage in total cost and calculates the amount of standard work's materials from the standard section and volume of I.L.M box girder bridge. Model (II) utilizes a correlation analysis (coefficient over 0.6 or more) between breakdown of standard works and input data that would be considered available information in preliminary design phase. Model(III) obtains conceptual estimating through multiple-regression analysis between the breakdown of standard works and all of input data related to them. In order to validate the clustering of coverage in the preliminary design phase, the variation of I.L.M cost coverage from multiple-regression analysis[model(III)] has been investigated which result in between -3.76% and 11.79%, comparing with AACE(Association for the Advancement of Cost Engineering) which informs its variation between -5% and +15% in the design phase. The model proposed from this research are envisioned to be improved to a great distinct if reliable cost date for P.S.C. girder bridges can be continually collected with reasonable accuracies.

• The Korean Journal of Applied Statistics
• /
• v.28 no.5
• /
• pp.921-932
• /
• 2015

This study analyzed military personnel survivability in regards to offensive operations according to the scientific military training data of a reinforced infantry battalion. Scientific battle training was conducted at the Korea Combat Training Center (KCTC) training facility and utilized scientific military training equipment that included MILES and the main exercise control system. The training audience freely engaged an OPFOR who is an expert at tactics and weapon systems. It provides a statistical analysis of data in regards to state-of-the-art military training because the scientific battle training system saves and utilizes all training zone data for analysis and after action review as well as offers training control during the training period. The methodologies used the Cox PH modeling (which does not require parametric distribution assumptions) and decision tree modeling for survival data such as CART, GUIDE, and CTREE for richer and easier interpretation. The variables that violate the PH assumption were stratified and analyzed. Since the Cox PH model result was not easy to interpret the period of service, additional interpretation was attempted through univariate local regression. CART, GUIDE, and CTREE formed different tree models which allow for various interpretations.

• Journal of Korea Water Resources Association
• /
• v.50 no.11
• /
• pp.745-758
• /
• 2017

The copula-based models have been successfully applied to hydrological modeling including drought frequency analysis and time series modeling. However, uncertainty estimation associated with the parameters of these model is not often properly addressed. In these context, the main purposes of this study are to develop the Bayesian inference scheme for bivariate copula functions. The main applications considered are two-fold: First, this study developed and tested an approach to copula model parameter estimation within a Bayesian framework for drought frequency analysis. The proposed modeling scheme was shown to correctly estimate model parameters and detect the underlying dependence structure of the assumed copula functions in the synthetic dataset. The model was then used to estimate the joint return period of the recent 2013~2015 drought events in the Han River watershed. The joint return period of the drought duration and drought severity was above 100 years for many of stations. The results obtained in the validation process showed that the proposed model could effectively reproduce the underlying distribution of observed extreme rainfalls as well as explicitly account for parameter uncertainty in the bivariate drought frequency analysis.

• Journal of Korean Society of Transportation
• /
• v.20 no.7
• /
• pp.177-183
• /
• 2002

Current state-of-the-art of VMS control is based upon simple knowledge-based inference engine with message set and each message's priority. And R&Ds of the VMS control are focused on the accurate detection and estimation of traffic condition of the subject roadways. However VMS display itself cannot achieve a desirable traffic allocation among alternative routes in the network In this context, VMS display strategy is the most crucial part in the VMS control. VMS itself has several limitations in its nature. It is generally known that VMS causes overreaction and concentration problems, which may be more serious in urban network than highway network because diversion should be more easily made in urban network. A feedback control algorithm is proposed in this paper to address the above-mentioned issues. It is generally true that feedback control approach requires low computational effort and is less sensitive to models inaccuracy and disturbance uncertainties. Major features of the proposed algorithm are as follows: Firstly, a regulator is designed to attain system optimal traffic allocation among alternative routes for each VMS in the network. Secondly, strategic messages should be prepared to realize the desirable traffic allocation, that is, output of the above regulator. VMS display strategy module is designed in this context. To evaluate Probable control benefit and to detect logical errors of the Proposed feedback algorithm, a offline simulation test is performed using real network in Daejon, Korea.

• Journal of the Korea Society of Computer and Information
• /
• v.26 no.4
• /
• pp.1-9
• /
• 2021

In this paper, we propose a training algorithm of support vector machine (SVM) with a sensitive variable. Although machine learning models enable automatic decision making in the real world applications, regulations prohibit sensitive information from being used to protect privacy. In particular, the privacy protection of the legally protected attributes such as race, gender, and disability is compulsory. We present an efficient least square SVM (LSSVM) training algorithm using a fully homomorphic encryption (FHE) to protect a partial sensitive attribute. Our framework posits that data owner has both non-sensitive attributes and a sensitive attribute while machine learning service provider (MLSP) can get non-sensitive attributes and an encrypted sensitive attribute. As a result, data owner can obtain the encrypted model parameters without exposing their sensitive information to MLSP. In the inference phase, both non-sensitive attributes and a sensitive attribute are encrypted, and all computations should be conducted on encrypted domain. Through the experiments on real data, we identify that our proposed method enables to implement privacy-preserving sensitive LSSVM with FHE that has comparable performance with the original LSSVM algorithm. In addition, we demonstrate that the efficient sensitive LSSVM with FHE significantly improves the computational cost with a small degradation of performance.

• KIPS Transactions on Computer and Communication Systems
• /
• v.9 no.12
• /
• pp.291-306
• /
• 2020

Nowadays, Data-Network-AI (DNA)-based intelligent services and applications have become a reality to provide a new dimension of services that improve the quality of life and productivity of businesses. Artificial intelligence (AI) can enhance the value of IoT data (data collected by IoT devices). The internet of things (IoT) promotes the learning and intelligence capability of AI. To extract insights from massive volume IoT data in real-time using deep learning, processing capability needs to happen in the IoT end devices where data is generated. However, deep learning requires a significant number of computational resources that may not be available at the IoT end devices. Such problems have been addressed by transporting bulks of data from the IoT end devices to the cloud datacenters for processing. But transferring IoT big data to the cloud incurs prohibitively high transmission delay and privacy issues which are a major concern. Edge computing, where distributed computing nodes are placed close to the IoT end devices, is a viable solution to meet the high computation and low-latency requirements and to preserve the privacy of users. This paper provides a comprehensive review of the current state of leveraging deep learning within edge computing to unleash the potential of IoT big data generated from IoT end devices. We believe that the revision will have a contribution to the development of DNA-based intelligent services and applications. It describes the different distributed training and inference architectures of deep learning models across multiple nodes of the edge computing platform. It also provides the different privacy-preserving approaches of deep learning on the edge computing environment and the various application domains where deep learning on the network edge can be useful. Finally, it discusses open issues and challenges leveraging deep learning within edge computing.

• Journal of the Korea Society of Computer and Information
• /
• v.27 no.12
• /
• pp.59-68
• /
• 2022

This paper presents a method for 1:1 verification by comparing the similarity between the given real product image and the drawing image. The proposed method combines two existing CNN-based deep learning models to construct a Siamese Network. After extracting the feature vector of the image through the FC (Fully Connected) Layer of each network and comparing the similarity, if the real product image and the drawing image (front view, left and right side view, top view, etc) are the same product, the similarity is set to 1 for learning and, if it is a different product, the similarity is set to 0. The test (inference) model is a deep learning model that queries the real product image and the drawing image in pairs to determine whether the pair is the same product or not. In the proposed model, through a comparison of the similarity between the real product image and the drawing image, if the similarity is greater than or equal to a threshold value (Threshold: 0.5), it is determined that the product is the same, and if it is less than or equal to, it is determined that the product is a different product. The proposed model showed an accuracy of about 71.8% for a query to a product (positive: positive) with the same drawing as the real product, and an accuracy of about 83.1% for a query to a different product (positive: negative). In the future, we plan to conduct a study to improve the matching accuracy between the real product image and the drawing image by combining the parameter optimization study with the proposed model and adding processes such as data purification.