• Annual Conference on Human and Language Technology
• /
• 2019.10a
• /
• pp.522-526
• /
• 2019

추출 요약(Extractive summarization)은 문서내에 주요한 요약정보가 되는 문장 또는 단어를 추출하여 요약을 생성하는 기법이다. 딥러닝 기법들이 많이 발전하면서 요약 기법에도 sequence-to-sequence와 같은 많은 시도들이 있었지만 대부분의 방법론들은 딥러닝의 모델 구조관점으로 접근하거나 요약에 있어서 단순히 입력 텍스트를 넣고 알고리즘이 처리하는 머신 리딩(Machine reading)관점으로 접근한다. 텍스트 요약 태스크 자체는 사람이 텍스트에 대한 정보 파악을 요약문을 통해 빠르게 하고 싶은 궁극적인 목표가 있으므로, 사람이 텍스트 요약에 필요한 인지처리과정을 반영할 필요가 있다. 결국, 기존의 머신 리딩보다는 휴먼 리딩(Human reading)에 관한 이해와 구조적 접근이 필요하다. 따라서 본 연구는 휴먼 리딩을 위한 인지처리과정을 위해 아이트래킹 데이터 기반의 새로운 추출 요약 모델을 제안한다.

• Journal of the Korea Society of Computer and Information
• /
• v.15 no.5
• /
• pp.33-38
• /
• 2010

In this paper, we proposed an algorithm implementing mouse functions using hand motion and number of fingers which are extracted from an image sequence. The sequence is acquired through a web camera and processed with image processing algorithms. The sequence is first converted from RGB model to YCbCr model to efficiently extract skin area and the extracted area is further processed using labeling, opening, and closing operations to decide the center of a hand. Based on the center position, the number of fingers is decided, which serves as the information to decide and perform a mouse function. Experimental results show that 94.0% of pointer moves and 96.0% of finger extractions are successful, which opens the possibility of further development for a commercial product.

• Journal of the Computational Structural Engineering Institute of Korea
• /
• v.26 no.1
• /
• pp.89-97
• /
• 2013

With regard to complex-shaped tall buildings whose plans and constructions have been gradually on the increase, this study was aimed to analyze their structural behaviors during construction by applications of construction sequences analyses to prototype models. For twisted tall buildings, total 18 models of with three conditions of a lateral load-resisting system, a twisting angle, and a construction method were selected. A diagrid system and a braced tube system were applied as a lateral load-resisting system. For each lateral load-resisting system, three types of plan with $0^{\circ}$, $1^{\circ}$, and $2^{\circ}$ twisting angles and three construction methods with construction sequences of exterior tube and interior frame were assumed. The structural performances of tall buildings under constructions were analyzed with results of lateral displacements from construction sequence analyses. Also, construction performances of the construction period and the maximum lift weight were compared.

• Journal of the Korean Society for Aeronautical & Space Sciences
• /
• v.48 no.6
• /
• pp.411-418
• /
• 2020

In this paper, approximation and optimization methods are proposed for the structural performance of the composite fan blade. Using these methods, we perform the optimal design of the stacking sequence to maximize stiffnesses without changing the mass and the geometric shape of the composite fan blade. In this study, the lamination parameters are introduced to reduce the design variables and space. From the characteristics of lamination parameters, we generate response surface model having a high fitness value. Considering the requirements of the optimal stacking sequence, the multi-objective optimization problem is formulated. We apply the two-step optimization method that combines gradient-based method and genetic algorithm for efficient search of an optimal solution. Finally, the finite element analysis results of the initial and the optimized model are compared to validate the approximation and optimization methods based on the lamination parameters.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.19 no.2
• /
• pp.93-103
• /
• 2009

In this paper, we propose a USIM-based Authentication Scheme for 3GPP Network Access. The proposed scheme improves the problems of existing authentication protocol in 3GPP Network such as sequence number synchronization problem, the storage overhead of authentication data, and bandwidth consumption between Serving Network and Home Network. Our proposal is based on the USIM-based Authentication and Key Agreement Protocol that is defined in 3GPP Specification. In our scheme, mobile nodes share a SK with Serving Network and use a time stamp when mobile nodes are performing an authentication procedure with Serving Network. By using time stamp, there is no reason for using sequence number to match the authentication vector between mobile nodes and networks. So, synchronization problem can be solved in our scheme. As well as our scheme uses an authentication vector, the storage overhead of authentication data in Serving Network and bandwidth consumption between networks can be improved.

• Journal of the Korean earth science society
• /
• v.33 no.3
• /
• pp.242-258
• /
• 2012

The Yellow Sea has sensitively responded to high-amplitude sea-level fluctuations during the late Quaternary. The repeated inundation and exposure have produced distinct transgression-regression successions with extensive exposure surfaces in Kyeonggi Bay. The late Quaternary strata consist of four seismic stratigraphic units, considered as depositional sequences (DS-1, DS-2, DS-3, and DS-4). DS-1 was interpreted as ridge-forming sediments of tidal-flat and estuarine channel-fill facies, formed during the Holocene highstand. DS-2 consists of shallow-marine facies in offshore area, which was formed during the regression of Marine Isotope Stage (MIS)-3 period. DS-3 comprises the lower transgressive facies and the upper highstand tidal-flat facies in proximal ridges and forced regression facies in distal ridges and offshore area. The lowermost DS-4 rests on acoustic basement rocks, considered as the shallow-marine and shelf deposits formed before the MIS-6 lowstand. This study suggests six depositional stages. During the first stage-A, MIS-6 lowstand, the Yellow Sea shelf was subaerially exposed with intensive fluvial incision and weathering. The subsequent rapid and high amplitude rise of sea level in stage-B until the MIS-5e highstand produced transgressive deposits in the lowermost part of the MIS-5 sequence, and the successive regression during the MIS-5d to -5a and the MIS-4 lowstand formed the upperpart of the MIS-5 sequence in stage-C. During the stage-D, from the MIS-4 lowstand to MIS-3c highstand period, the transgressive MIS-3 sequence formed in a subtidal environment characterized by repetitive fluvial incision and channel-fill deposition in exposed area. The subsequent sea-level fall culminating the last glacial maximum (Stage-E) made shallow-marine regressive deposits of MIS-3 sequence in offshore distal area, whereas it formed fluvial channel-fills and floodplain deposits in the proximal area. After the last glacial maximum, the overall Yellow Sea shelf was inundated by the Holocene transgression and highstand (Stage-F), forming the Holocene transgressive shelf sands and tidal ridges.

• Geophysics and Geophysical Exploration
• /
• v.25 no.2
• /
• pp.59-70
• /
• 2022

Recent studies demonstrate that machine learning has expanded in the field of seismic interpretation. Many convolutional neural networks have been developed for seismic sequence identification, which is important for seismic interpretation. However, expense and time limitations indicate that there is insufficient data available to provide a sufficient dataset to train supervised machine learning programs to identify seismic sequences. In this study, patch division and data augmentation are applied to mitigate this lack of data. Furthermore, to obtain spatial information that could be lost during patch division, an artificial channel is added to the original data to indicate depth. Seismic sequence identification is performed using a U-Net network and the Netherlands F3 block dataset from the dGB Open Seismic Repository, which offers datasets for machine learning, and the predicted results are evaluated. The results show that patch-based U-Net seismic sequence identification is improved by data augmentation and the addition of an artificial channel.

• Transactions of the Korean Society of Mechanical Engineers A
• /
• v.20 no.2
• /
• pp.431-439
• /
• 1996

New design principles nad necessary conditions for a mechanical system have been suggested to be kept in the design process using a washing machine transmission as a model. The necessary conditions are funcitnal requirement condition and spatial arrangement condition. The design principles to satisfy the necessary conditions are the principle of sequence and the principle of expansion. Decision sequence for state variables and design varibles of various mechanicla elements have been formulated. New automatic design program for washing machine transmission has been developed observing the necessary conditions and design principles investigated in this study. It was verified to be very effective to follow the design conditions, principles nad formulated decision sequence in mechanical system design process.

• The Transactions of the Korean Institute of Power Electronics
• /
• v.20 no.3
• /
• pp.273-279
• /
• 2015

A zero vector selection method to reduce switching losses for model predictive control (MPC) of voltage source inverter is proposed. A conventional MPC of voltage source inverter has not been proposed, and a method to select the redundancy of the zero vector is required for this study. In this paper, the redundancy of the zero vectors is selected with generating a zero sequence voltage to reduce switching losses. The zero vector of 2-level inverter is determined by determining sign of the zero sequence voltage. In the proposed method, the quality of the current is retained and switching loss can be reduced compared with the conventional method. This result was verified by P-sim simulation and experiments.

• KIPS Transactions on Software and Data Engineering
• /
• v.11 no.6
• /
• pp.245-254
• /
• 2022

Proteins are the basic unit of all life activities, and understanding them is essential for studying life phenomena. Since the emergence of the machine learning methodology using artificial neural networks, many researchers have tried to predict the function of proteins using only protein sequences. Many combinations of deep learning models have been reported to academia, but the methods are different and there is no formal methodology, and they are tailored to different data, so there has never been a direct comparative analysis of which algorithms are more suitable for handling protein data. In this paper, the single model performance of each algorithm was compared and evaluated based on accuracy and speed by applying the same data to CNN, LSTM, and GRU models, which are the most frequently used representative algorithms in the convergence research field of predicting protein functions, and the final evaluation scale is presented as Micro Precision, Recall, and F1-score. The combined models CNN-LSTM and CNN-GRU models also were evaluated in the same way. Through this study, it was confirmed that the performance of LSTM as a single model is good in simple classification problems, overlapping CNN was suitable as a single model in complex classification problems, and the CNN-LSTM was relatively better as a combination model.