• 정보처리학회논문지:소프트웨어 및 데이터공학
• /
• 제11권3호
• /
• pp.133-140
• /
• 2022

우리는 유학생이 작성한 한국어 쓰기 답안지의 점수 구간을 예측하는 문제에서 세 개의 딥러닝 기반 한국어 언어모델의 예측 성능을 조사한다. 이를 위해 총 304편의 답안지로 구성된 실험 데이터 세트를 구축하였는데, 답안지의 주제는 직업 선택의 기준('직업'), 행복한 삶의 조건('행복'), 돈과 행복('경제'), 성공의 정의('성공')로 다양하다. 이들 답안지는 네 개의 점수 구간으로 구분되어 평어 레이블(A, B, C, D)이 매겨졌고, 총 11건의 점수 구간 예측 실험이 시행되었다. 구체적으로는 5개의 '직업' 답안지 점수 구간(평어) 예측 실험, 5개의 '행복' 답안지 점수 구간 예측 실험, 1개의 혼합 답안지 점수 구간 예측 실험이 시행되었다. 이들 실험에서 세 개의 딥러닝 기반 한국어 언어모델(KoBERT, KcBERT, KR-BERT)이 다양한 훈련 데이터로 미세조정되었다. 또 두 개의 전통적인 확률적 기계학습 분류기(나이브 베이즈와 로지스틱 회귀)도 그 성능이 분석되었다. 실험 결과 딥러닝 기반 한국어 언어모델이 전통적인 기계학습 분류기보다 우수한 성능을 보였으며, 특히 KR-BERT는 전반적인 평균 예측 정확도가 55.83%로 가장 우수한 성능을 보였다. 그 다음은 KcBERT(55.77%)였고 KoBERT(54.91%)가 뒤를 이었다. 나이브 베이즈와 로지스틱 회귀 분류기의 성능은 각각 52.52%와 50.28%였다. 학습된 분류기 모두 훈련 데이터의 부족과 데이터 분포의 불균형 때문에 예측 성능이 별로 높지 않았고, 분류기의 어휘가 글쓰기 답안지의 오류를 제대로 포착하지 못하는 한계가 있었다. 이 두 가지 한계를 극복하면 분류기의 성능이 향상될 것으로 보인다.

• Asian-Australasian Journal of Animal Sciences
• /
• 제14권2호
• /
• pp.263-267
• /
• 2001

Beef samples of loin eye area from New Zealand, USA and three quality grades of Hanwoo were analyzed using near infrared spectrophotometer with reference values from laboratory optical Chromameter to determine effective spectrum range and mathematical treatment for determination of color values. $R^2s$ of prediction models were not improved much by calibrating with whole light range (400~2500 nm) compared to using visible range (400~1100 nm). Standard errors of calibration and prediction were influenced by possible bias due to sampling non-homogeneous sample sources. However, partial differentiation in the first order was more stable against sampling biases than second derivatives of the spectra. Lightness value was little different among the five sample sources of beef. Beef samples from USA were brighter and more reddish than beefs of Hanwoo or from New Zealand (p<0.05). Yellowness of USA beef was the highest followed by beef from New Zealand, which was also higher than Hanwoo beefs of three quality grades (p<0.05).

• ETRI Journal
• /
• 제22권4호
• /
• pp.32-39
• /
• 2000

In low earth orbit (LEO) satellite communication systems, more severe phase distortion due to Doppler shift is frequently detected in the received signal than in cases of geostationary earth orbit (GEO) satellite systems or terrestrial mobile systems. Therefore, an estimation of Doppler shift would be one of the most important factors to enhance performance of LEO satellite communication system. In this paper, a new adaptive Doppler compensation scheme using location information of a user terminal and satellite, as well as a weighting factor for the reduction of prediction error is proposed. The prediction performance of the proposed scheme is simulated in terms of the prediction accuracy and the cumulative density function of the prediction error, with considering the offset variation range of the initial input parameters in LEO satellite system. The simulation results showed that the proposed adaptive compensation algorithm has the better performance accuracy than Ali's method. From the simulation results, it is concluded the adaptive compensation algorithm is the most applicable method that can be applied to LEO satellite systems of a range of altitude between 1,000 km and 2,000 km for the general error tolerance level, M = 250 Hz.

• 전기학회논문지
• /
• 제64권1호
• /
• pp.107-112
• /
• 2015

This paper introduces an evolutionary nonlinear regression based compensation technique for the short-range prediction of wind speed using AWS(Automatic Weather Station) data. Development of an efficient MOS(Model Output Statistics) is necessary to correct systematic errors of the model, but a linear regression based MOS is hard to manage an irregular nature of weather prediction. In order to solve the problem, a nonlinear and symbolic regression method using GP(Genetic Programming) is suggested for a development of MOS wind forecast guidance. Also FCM(Fuzzy C-Means) clustering is adopted to mitigate bias of wind speed data. The purpose of this study is to evaluate the accuracy of the estimation by a GP based nonlinear MOS for 3 days prediction of wind speed in South Korean regions. This method is then compared to the UM model and has shown superior results. Data for 2007-2009, 2011 is used for training, and 2012 is used for testing.

• Asian-Australasian Journal of Animal Sciences
• /
• 제18권7호
• /
• pp.1036-1040
• /
• 2005

Near-infrared reflectance spectroscopic (NIRS) calibrations were developed for accurate and fast prediction of whole broiler chicken carcass composition. The Feed and Forage Foss systems Model 5000 Reflectance Transport Model 5000 with near-infrared reflectance spectroscopy (NIRS)-WinISI II windows software was used for this purpose. One equation was developed for the prediction of each carcass component. One hundred and fifty freeze dried broiler whole carcass samples were ground in a Cyclotech 1,093 sample mill and analyzed for dry matter, protein, fat, calcium and phosphate. Samples were divided into two sets: a calibration set from which equations were derived and a prediction set used to validate these equations. The chemical analysis values (mean${\pm}$SD) were calculated based on dry matter basis as follows: dry matter: 33.41${\pm}$2.78 (range: 26.41-43.47), protein: 54.04${\pm}$6.63 (range: 36.20-76.09), fat 35.44${\pm}$8.34 (range: 7.50-55.03), calcium 2.55${\pm}$0.65 (range: 0.99-4.41), phosphorus 1.38${\pm}$0.26 (range: 0.60-2.28). One hundred and three samples were used to calibrate the equations and prediction values. The software used was modified to obtain partial least square regression statistics, as it is the most suitable for natural products analysis. The coefficients of determination ($R^2$) and the standard errors of prediction were 0.82 and 1.83 for the dry matter, 0.96 and 1.98 for protein, 0.99 and 1.07 for fat, 0.90 and 0.30 for calcium and 0.91 and 0.11 for phosphorus, respectively. The present study indicated that NIRS can be calibrated to predict the whole broiler carcass chemical composition, including minerals in a rapid, accurate, and cost effective manner. It neither requires skilled operators nor generates hazardous waste. These findings may have practical importance to improve instrumental procedures for quick evaluation of broiler carcass composition.

• 한국수자원학회논문집
• /
• 제54권7호
• /
• pp.485-493
• /
• 2021

지하수는 지표수와 함께 용수로 사용가능한 중요한 수자원이며 특히 섬 지역의 경우 전체 수자원 중 지하수의 이용 비율이 상대적으로 높기 때문에 안정적인 이용을 위해 지하수위 변동성에 대한 연구는 필수적이다. 지하수위 변동성의 예측 및 분석을 위해 인공지능 모델을 활용한 연구들이 지속적으로 증가하고 있으나 지하수위 예측결과의 적절성을 판단할 수 있는 평가기준을 제시한 연구는 충분하지 않다. 본 연구에서는 허용가능한 지하수위 예측오차의 범위를 제시하기 위해 과거 20년 동안 전 세계 다양한 지역을 대상으로 인공지능 모델을 활용하여 지하수위를 예측한 연구결과들을 종합적으로 분석하였다. 그 결과 관측지하수위의 변동성이 커질수록 인공지능 모델에 의한 지하수위 예측오차는 증가하였다. 따라서 관측지하수위 최대변동폭과 예측오차 간의 상관성과 기존 연구들에서 제시한 평가지수들을 고려하여 평가기준을 산정하였으며, 인공지능 모델에 의한 지하수위 예측결과의 적절한 평가기준은 도출된 선형회귀식에 의한 평균제곱근오차 또는 최대오차 이하이거나, NSE ≥ 0.849 또는 R² ≥ 0.880 이다. 이 허용가능한 오차범위는 인공지능 모델을 활용한 지하수위 예측결과의 적절성 판단을 위한 참고자료로 사용할 수 있다.

• 한국정보통신학회:학술대회논문집
• /
• 한국정보통신학회 2017년도 추계학술대회
• /
• pp.159-161
• /
• 2017

본 논문에서는 고성능 HEVC 부호기를 위한 적응적 탐색영역 할당과 제안하는 알고리즘에 적합한 하드웨어 구조를 제안한다. 기존 움직임 벡터는 예측 성능을 향상하기 위하여 주변 블록의 움직임 벡터들을 예측 벡터 후보로 구성하고 현재 움직임 벡터와 최소의 차이를 가지는 하나의 움직임 벡터를 이용하여 일정한 크기의 탐색영역을 할당한다. 제안하는 알고리즘은 주변 네 개의 블록에 대한 움직임 벡터들의 구조에 따라 탐색영역의 크기를 직사각형과 옥타곤 형태로 할당함으로써 탐색영역의 크기를 축소하여 연산시간을 감소시켰다. 또한, 네 개의 움직임 벡터들을 모두 사용함에 따라 더 정확한 예측이 가능하며, 하드웨어에 적합한 형태로 구현함으로써 하드웨어 면적 및 연산시간을 효과적으로 감소시켰다.

• 대기
• /
• 제25권4호
• /
• pp.623-637
• /
• 2015

The accurate analysis of water vapor in initial of numerical weather prediction (NWP) model is required as one of the necessary conditions for the improvement of heavy rainfall prediction and reduction of spin-up time on a very-short-range forecast. To study this effect, the impact of a ground-based Global Positioning System (GPS)-Precipitable Water Vapor (PWV) on very-short-range forecast are examined. Data assimilation experiments of GPS-PWV data from 19 sites over the Korean Peninsula were conducted with Advanced Storm-scale Analysis and Prediction System (ASAPS) based on the Korea Meteorological Administration's Korea Local Analysis and Prediction System (KLAPS) included "Hot Start" as very-short-range forecast system. The GPS total water vapor was used as constraint for integrated water vapor in a variational humidity analysis in KLAPS. Two simulations of heavy rainfall events show that the precipitation forecast have improved in terms of ETS score compared to the simulation without GPS-PWV data. In the first case, the ETS for 0.5 mm of rainfall accumulated during 3 hrs over the Seoul-Gyeonggi area shows an improvement of 0.059 for initial forecast time. In other cases, the ETS improved 0.082 for late forecast time. According to a qualitative analysis, the assimilation of GPS-PWV improved on the intensity of precipitation in the strong rain band, and reduced overestimated small amounts of precipitation on the out of rain band. In the case of heavy rainfall during the rainy season in Gyeonggi province, 8 mm accompanied by the typhoon in the case was shown to increase to 15 mm of precipitation in the southern metropolitan area. The GPS-PWV assimilation was extremely beneficial to improving the initial moisture analysis and heavy rainfall forecast within 3 hrs. The GPS-PWV data on variational data assimilation have provided more useful information to improve the predictability of precipitation for very short range forecasts.

• 한국소음진동공학회:학술대회논문집
• /
• 한국소음진동공학회 2007년도 춘계학술대회논문집
• /
• pp.56-60
• /
• 2007

Nowadays shear wall structural system is gradually changing to framed structure. For this reason, lightweight panel is increasingly being used as separating walls. One of design methods to obtain high transmission loss is double panel. To predict the acoustic performance of double panel, prediction of transmission loss of single panel must be performed, previously. In this study, the predicted values for four single panels were compared with the measured values. The result shows the arithmetical average deviations(100Hz to 3150Hz) between the predicted and measured transmission loss were in range between 1.1dB and 3.9dB. The predicted values were generally lower than measured values above critical frequency. The single-number quantities, $R_W+C$, were predicted in range between 36dB to 38dB, and the differences of single-number quantities between the predicted and measured value were within 1dB.

• Bulletin of the Korean Chemical Society
• /
• 제28권9호
• /
• pp.1477-1484
• /
• 2007

Artificial neural networks (ANNs), for a first time, were successfully developed for the prediction partial molar heat capacity of aqueous solutions at infinite dilution for various polar aromatic compounds over wide range of temperatures (303.55-623.20 K) and pressures (0.1-30.2 MPa). Two three-layered feed forward ANNs with back-propagation of error were generated using three (the heat capacity in T = 303.55 K and P = 0.1 MPa, temperature and pressure) and six parameters (four theoretical descriptors, temperature and pressure) as inputs and its output is partial molar heat capacity at infinite dilution. It was found that properly selected and trained neural networks could fairly represent dependence of the heat capacity on the molecular descriptors, temperature and pressure. Mean percentage deviations (MPD) for prediction set by the models are 4.755 and 4.642, respectively.