• Annual Conference on Human and Language Technology
• /
• 2023.10a
• /
• pp.203-208
• /
• 2023

본 논문에서는 대규모 언어 모델의 인컨텍스트 러닝과 프롬프팅을 활용하여 수치 추론 태스크 데이터셋을 효과적으로 증강시킬 수 있는 방법론을 제안한다. 또한 모델로 하여금 수치 추론 데이터의 이해를 도울 수 있는 전처리와 요구사항을 만족하지 못하는 결과물을 필터링 하는 검증 단계를 추가하여 생성되는 데이터의 퀄리티를 보장하고자 하였다. 이렇게 얻어진 증강 절차를 거쳐 증강을 진행한 뒤 추론용 모델 학습을 통해 다른 증강 방법론보다 우리의 방법론으로 증강된 데이터셋으로 학습된 모델이 더 높은 성능을 낼 수 있음을 보였다. 실험 결과 우리의 증강 데이터로 학습된 모델은 원본 데이터로 학습된 모델보다 모든 지표에서 2%p 이상의 성능 향상을 보였으며 다양한 케이스를 통해 우리의 모델이 수치 추론 학습 데이터의 다양성을 크게 향상시킬 수 있음을 확인하였다.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.19 no.3
• /
• pp.173-178
• /
• 2019

To analyze the diffraction properties of optical signals by periodic blazed 2D diffraction gratings, Toeplitz dielectric tensor is first defined and formulated by 2D spatial Fourier expansions associated with asymmetric blazed grating profile. The characteristic modes in each layer is then based on eigenvalue problem, and the complete solution is found rigorously in terms of modal transmission-line theory (MTLT) to address the pertinent boundary-value problems. Toeplitz matrix of symmetric and sawtooth profiles is derived from that of asymmetric blazed grating profile, and the diffraction properties for each profile are numerically simulated. The numerical results reveal that the asymmetric and symmetric profiles behave as anti-reflection GMR filter while the sawtooth profile works better as anti-transmission one rather than anti-reflection filter.

• Journal of the Korean Geosynthetics Society
• /
• v.19 no.4
• /
• pp.85-93
• /
• 2020

This study describes the results of laboratory model test on settlement of soil contaminated with heavy metals, in order to evaluate the applicability of VMS to the measurement of gound settlements generated during the purification of contaminated soil. The measurement results for settlement of contaminated soil were compared using a 3D-Visual Monitoring System (VMS) based on digital photogrammetry and a total station. The test result showed that the settlement of the soil contaminated with heavy metals occurred a lot in the experimental condition in which the hydrophilic filter was applied. The minimum and maximum error ranges of VMS were calculated as ±0.36mm and ±0.87mm, respectively, and the error of VMS was satisfied in all experimental conditions. The average error rate of VMS was lower in the hydrophilic filter condition than in the hydrophobic filter condition. Therefore, it was evaluated that VMS can be applied to measure the settlement of contaminated soil.

• The Journal of Korean Institute of Electromagnetic Engineering and Science
• /
• v.19 no.4
• /
• pp.418-426
• /
• 2008

In this paper, investigations on the impedance characteristics of a DGS(Defected Ground Structure) slot in the groud plane of microstripline are presented in spectral domain and applied to the characteristic improvement of stepped impedance microstrip low pass filter(LPF). In this method, expressions for the impedance of a DGS slot are derived from self-reaction of the angular spectrum of plane waves and the discontinuity in the modal voltage. The numerical results are compared with those of the rigorous full-wave method and are shown to produce reasonably accurate data. And the stepped impedance microstrip low pass filter is designed and fabricated with the uniform and nonuniform DGS slots for improving the frequency responses. The experiments show that the proposed filter with slots in the ground plane has a wider stopband and sharper cutoff response.

• Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.29 no.6
• /
• pp.641-650
• /
• 2011

Airborne LIDAR system, utilized in various forest studies, provides efficiently spatial information about vertical structures of forest areas. The tree height is one of the most essential measurements to derive forest information such as biomass, which can be estimated from the forest terrain model. As the terrain model is generated by the interpolation of ground points extracted from LIDAR data, filtering methods with high reliability to classify reliably the ground points are required. In this paper, we applied three representative filtering methods to forest LIDAR data with diverse characteristics, measured the errors and performance of these methods, and analyzed the causes of the errors. Based on their complementary characteristics derived from the analysis results, we have attempted to combine the results and checked the performance improvement. In most test areas, the convergence method showed the satisfactory results, where the filtering performance were improved more than 10% in maximum. Also, we have generated DTM using the classified ground points and compared with the verification data. The DTM retains about 17cm RMSE, which can be sufficiently utilized for the derivation of forest information.

• Journal of the Korean Society for Aeronautical & Space Sciences
• /
• v.40 no.4
• /
• pp.292-299
• /
• 2012

The line-of-sight(LOS) rate is estimated using Kalman filter in Radio-Frequency(RF) seeker. For the two axis gimbaled seeker, proper system modeling is considered and the basic filter structure is set up. The main issue for Kalman filter is choosing the proper process and measurement noise. For the measurement process, the signal to noise ratio(SNR) and other components are introduced. To cope with the eclipse problem or other abnormal seeker status, the pseudo input signal concept is proposed. By conditioning abnormal signals, the LOS rate estimation performance is increased. The process noise is also an important factor in the propagation phase. The analytical approach on a process noise component is performed and a reliable region for the filter is calculated based on the eigenvalue analysis. Some numerical simulations are performed to check the validity of suggested algorithm.

• Journal of the Korean earth science society
• /
• v.38 no.1
• /
• pp.11-23
• /
• 2017

The short term predictability of wind resources is an important factor in evaluating the economic feasibility of a wind power plant. As a method of improving the predictability, a Bayesian Kalman filter is applied as the model data postprocessing. At this time, a statistical training period is needed to evaluate the correlation between estimated model and observation data for several Kalman training periods. This study was quantitatively analyzes for the prediction characteristics according to different training periods. The prediction of the temperature and wind speed with 3-day short term Bayesian Kalman training at Taebaek area is more reasonable than that in applying the other training periods. In contrast, it may produce a good prediction result in Ieodo when applying the training period for more than six days. The prediction performance of a Bayesian Kalman filter is clearly improved in the case in which the Weather Research Forecast (WRF) model prediction performance is poor. On the other hand, the performance improvement of the WRF prediction is weak at the accurate point.

• Journal of the Korea Society of Computer and Information
• /
• v.18 no.6
• /
• pp.9-19
• /
• 2013

This paper deals with fuzzy $H_{\infty}$ filtering problem of discrete-time nonlinear Markovian jump systems with state and output time delays. The purpose is to design fuzzy $H_{\infty}$ filter such that the corresponding estimation error system with time delays and initial state uncertainties is stochastically stable and satisfies an $H_{\infty}$ performance level. A sufficient condition for the existence of fuzzy $H_{\infty}$ filter is given in terms of matrix inequalities. In order to relax conservatism, a stochastic mode dependent fuzzy Lyapunov function is employed. The Lyapunov function not only is dependent on the operation modes of system, but also includes the fuzzy membership functions. An illustrative example is finally given to show the applicability and effectiveness of the proposed method.

• Journal of the Korean Society for Aeronautical & Space Sciences
• /
• v.50 no.8
• /
• pp.541-550
• /
• 2022

An effective real time fault diagnosis approach for UAV engine is drawn from IMM filter and GLRT methods. For this purpose based on the linear diagnosis model derived from engine dynamic performance analysis the Kalman filter for residual estimation and each method are applied to the fault diagosis of the actuator for engine control sensors. From the process of the IMM filter application the effective FDI measure is obtained and the state responses due to actuator fault are estimated. Likewise from the GLRT method the fault magnitudes of actuator and sensors are estimated associated with some FDI functionings. The numerical simulations verify the effectiveness of the IMM filter for FDI and the GLRT in estimating the fault magnitudes of each fault mode.

• Journal of the Computational Structural Engineering Institute of Korea
• /
• v.26 no.4
• /
• pp.241-246
• /
• 2013

From the numerical results of density-based topology design optimization, a CAD geometric model is constructed and fabricated using 3D printer to experimentally validate the optimal design. In the process of topology design optimization, we often experience checkerboard phenomenon and complicated branches, which could result in the manufacturing difficulty of the obtained optimal design. Sensitivity filtering and morphology methods are used to resolve the aforementioned issues. Identical volume fraction is used in both numerical and experimental models for precise validation. Through the experimental comparison of stiffness in various designs including the optimal design, it turns out that the optimal design has the highest stiffness and the experimental result of compliance matches very well with the numerical one.