• Proceedings of the Korea Water Resources Association Conference
• /
• 2015.05a
• /
• pp.453-453
• /
• 2015

미계측 유역에서의 수문 예측은 유역의 다차원 시공간에서 일어나고 있는 수문학적 기능에 대한 깊은 이해와 성찰을 요구한다. 유역면적의 2/3가 미계측 지역인 임진강 유역은 북한지역과 중첩하고 있어 관측자료가 불충분 하고 소량의 관측자료가 존재하더라도 기후변화로 인해 환경이 변화하기 때문에 미계측 지역 연구에 적당하다. 이에 따라 수문학적 반응을 예측할 수 있도록 비접촉 비파괴적인 도구를 이용하여 수집된 자료를 통해 미계측 유역에 대한 정확한 신뢰성 구축을 마련할 필요가 있다고 판단된다. 신뢰성 구축을 위한 방법으로는 현장답사 및 항공사진에 비하여 넓은 지역을 한번에 관측할 수 있는 Landsat TM 영상을 이용하여 북한의 지형과 토지피복특성 등을 구축하고, 준분포형 모형인 SLURP를 이용하여 소유역으로 구분 된 ASA의 하도 추적을 통해 전체유역의 출구지점 유출량을 산정하였다. 또한 예측 불확실성을 감소시키기 위해 wamis에서 제공하는 GIS Data와 위성영상의 Data를 비교하여 분석하였다. 그 결과 미계측 지역의 불확실성을 최소화 시킬 수 있는 비교 분석이 가능하였다.

• Journal of the Institute of Electronics Engineers of Korea TC
• /
• v.46 no.11
• /
• pp.18-24
• /
• 2009

This paper proposes two novel prediction-based adaptive selection cooperation schemes combined with a new relay selection strategy. In the proposed schemes, the destination predicts whether the transmission will be successful or not before a single relay is selected to transmit source's decoded data. Depending on the prediction, the destination feeds back a command to the whole network. Numerical results show that the proposed schemes combined with the relay selection strategy successfully reduce its outage probability, improve its throughput, save transmitted power, and prolong the lifetime of the network.

• Annual Conference on Human and Language Technology
• /
• 2019.10a
• /
• pp.18-22
• /
• 2019

FAQ 분류는 자주 묻는 질문을 범주화하고 사용자 질의에 대해 가장 유사한 클래스를 추론하는 방식으로 진행된다. FAQ 데이터셋은 클래스가 다수 존재하기 때문에 클래스 간 포함 및 연관 관계가 존재하고 특정 데이터가 서로 다른 클래스에 동시에 속할 수 있다는 특징이 있다. 그러나 최근 FAQ 분류는 다중 클래스 분류 방법론을 적용하는 데 그쳤고 FAQ 데이터셋의 특징을 모델에 반영하는 연구는 미미했다. 현 분류 방법론은 이러한 FAQ 데이터셋의 특징을 고려하지 못하기 때문에 정답으로 해석될 수 있는 예측도 오답으로 여기는 경우가 발생한다. 본 논문에서는 신뢰성이 부족한 FAQ 데이터셋에서도 분류를 잘 하기 위해 손실 함수를 조정하는 정규화 기법을 소개한다. 이 정규화 기법은 클래스 간 포함 및 연관 관계를 반영할 수 있도록 오답을 예측한 경우에도 예측 강도에 비례하여 손실을 줄인다. 이는 오답을 높은 확률로 예측할수록 데이터의 신뢰성이 낮을 가능성이 크다고 판단하여 학습을 강하게 하지 않게 하기 위함이다. 실험을 위해서는 다중 클래스 분류에서 가장 좋은 성능을 보이고 있는 모형인 BERT를 이용했으며, 비교 실험을 위한 정규화 방법으로는 통상적으로 사용되는 라벨 스무딩을 채택했다. 실험 결과, 본 연구에서 제안한 방법은 기존 방법보다 성능이 개선되고 보다 안정적으로 학습이 된다는 것을 확인했으며, 데이터의 신뢰성이 부족한 상황에서 효과적으로 분류를 수행함을 알 수 있었다.

• Proceedings of the Optical Society of Korea Conference
• /
• 2007.02a
• /
• pp.62-63
• /
• 2007

• Management & Information Systems Review
• /
• v.29 no.2
• /
• pp.1-25
• /
• 2010

This study attempted to empirically test the determinants of stock returns in Korean stock market applying multi-factor model proposed by Haugen and Baker(1996). Regression models were developed using 16 variables related to liquidity, risk, historical price, price level, and profitability as independent variables and 690 stock monthly returns as dependent variable. For the statistical analysis, the data were collected from the Kis Value database and the tests of forecasting power in this study minimized various possible bias discussed in the literature as possible. The statistical results indicated that: 1) Liquidity, one-month excess return, three-month excess return, PER, ROE, and volatility of total return affect stock returns simultaneously. 2) Liquidity, one-month excess return, three-month excess return, six-month excess return, PSR, PBR, ROE, and EPS have an antecedent influence on stock returns. Meanwhile, realized returns of decile portfolios increase in proportion to predicted returns. This results supported previous study by Haugen and Baker(1996) and indicated that firm-characteristic model can better predict stock returns than CAPM. 3) The firm-characteristic model has better predictive power than Fama-French three-factor model, which indicates that a portfolio constructed based on this model can achieve excess return. This study found that expected return factor models are accurate, which is consistent with other countries' results. There exists a surprising degree of commonality in the factors that are most important in determining the expected returns among different stocks.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.48 no.2
• /
• pp.1-10
• /
• 2011

In this paper, we propose the algorithm for the AR(Augmented Reality) system, which is robust to the brightness change of light. In the proposed method, the histogram specification is achieved using the sample histogram, obtained from the frames in which the target objects could be detected successful. And When the object key-points couldn't be detected by the displacement of camera positions, the positions of non-detected key-points ware estimated using the linear KF(Kalman Filter). When the proposed algorithm is applied in the AR systems, the object key-points can be detected three times as much as the existing others. In addition, to prove the more efficiency of the proposed algorithm, we implemented the AR game, and could know that the performance is the more advanced than the others. The proposed algorithm can be used for the AR environments, which high efficiency is required such as the AR game, or the implementation of AR systems which are robust to the change of lights, etc.

• The Korean Journal of Applied Statistics
• /
• v.28 no.3
• /
• pp.561-571
• /
• 2015

When we consider an optimal design to predict the response corresponding to the point outside the design region, we are extremely careful about choosing the design criteria for selecting the support points. The assumed model and its accompanying error structure should be assumed to extend beyond the design region for the selected design criteria to be valid. Thus, we modify the existing design criteria such as extrapolation-optimality to be suited to those situations. We propose some maximin approaches in this paper. Simple and quadratic regression models are tested to find the basic characteristics of such maximin approaches. Some main findings are discussed in the conclusion.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.14 no.8
• /
• pp.1918-1924
• /
• 2010

This In this paper, we propose an efficient algorithm for motion estimation of the image sequences with luminance fluctuation. For such sequences, conventional motion estimation methods based on the difference of pixel values usually produce the erroneous motion information. The proposed algorithm defines the luminance fluctuation as a linear model with gain and offset parameter, and extracts motion information using gradient and phase of the corresponding local region within consecutive frames. Therefor the method is robust to the luminance change of the frames. We test our algorithm for the ground truth sequence with artificially added luminance change and motion, and real sequences corrupted by the flicker. The results shows that the proposed algorithm outperforms the conventional methods.

• Journal of Internet Computing and Services
• /
• v.10 no.6
• /
• pp.81-98
• /
• 2009

Collaborative filtering, one of the most successful technologies among recommender systems, is a system assisting users in easily finding the useful information and supporting the decision making. However, despite of its success and popularity, one notable issue is incredibility of recommendations by unreliable users called shilling attacks. To deal with this problem, in this paper, we analyze the type of shilling attacks and propose a unique method of building a model for protecting the recommender system against manipulated ratings. In addition, we present a method of applying the model to collaborative filtering which is highly robust and stable to shilling attacks.

• The KIPS Transactions:PartB
• /
• v.14B no.7
• /
• pp.473-480
• /
• 2007

Feature tracking in video frame sequences has suffered from the instability and the frequent failure of feature matching between two successive frames. In this paper, we propose a robust 2D feature tracking method that is stable to long video sequences. To improve the stability of feature tracking, we predict the spatial movement in the current image frame using the state variables. The predicted current movement is used for the initialization of the search window. By computing the feature similarities in the search window, we refine the current feature positions. Then, the current feature states are updated. This tracking process is repeated for each input frame. To reduce false matches, the outlier rejection stage is also introduced. Experimental results from real video sequences showed that the proposed method performs stable feature tracking for long frame sequences.