• Proceedings of the Korea Information Processing Society Conference
• /
• 2023.05a
• /
• pp.476-478
• /
• 2023

딥러닝은 컴퓨터 비전의 상당한 발전을 기여했지만, 딥러닝 모델을 학습하려면 대규모 데이터 세트가 필요하다. 이를 해결하기 위해 데이터 증강 기술이 주목받고 있다. 본 논문에서는 객체 추출 바운딩 박스와 원본 이미지의 바운딩 박스를 결합하여 합성 데이터 생성기법을 제안한다. 원본 이미지와 동일한 범주의 데이터셋에서 참조 이미지의 객체를 추출한 다음 생성 모델을 사용하여 참조 이미지와 원본 이미지의 특징을 통합하여 새로운 합성 이미지를 만든다. 실험을 통해, 생성 기법을 통한 딥러닝 모델의 성능향상을 보여준다.

• Annual Conference on Human and Language Technology
• /
• 2023.10a
• /
• pp.19-24
• /
• 2023

자연어 처리 분야에서 데이터의 중요성이 더욱 강조되고 있으며, 특히 리소스가 부족한 도메인에서 데이터 부족 문제를 극복하는 방법으로 데이터 증강이 큰 주목을 받고 있다. 이 연구는 대규모 언어 모델(Large Language Model, LLM)을 활용한 키워드 기반 데이터 증강 방법을 제안하고자 한다. 구체적으로 한국어에 특화된 LLM을 활용하여 주어진 키워드를 기반으로 특정 주제에 관한 대화 내용을 생성하고, 이를 통해 대화 주제를 분류하는 분류 모델의 성능 향상을 입증했다. 이 연구 결과는 LLM을 활용한 데이터 증강의 유의미성을 입증하며, 리소스가 부족한 상황에서도 이를 활용할 수 있는 방법을 제시한다.

• The Journal of the Korea Contents Association
• /
• v.14 no.10
• /
• pp.472-485
• /
• 2014

This study focuses on the news making system of the prime time news of PSB in Korean society, where the presence of ethnic minorities is increasing rapidly. Although the World Wide Web has become one of the most attractive media over the last decade, Korean PSB, Korean Broadcasting System (KBS), still remains the most popular and influential medium. Therefore, the process of analyzing news making system of ethnic minorities in Korean society represented in Korean PSB as a social construction is meaningful in that it provides an important key to understand the cultural and political background and characteristics of society. For this purpose, the article tries to understand news making process when producing news related to ethnic minorities in the Korean society such as migrant workers, married migrant women and mixed-heritage children of multicultural families by interview with ten reporters in KBS. As a result, most KBS reporters had stereotypes towards multiculturalism and migrants and news reports relating to ethnic minorities are usually produced routinely, using a set of rules that have become part of KBS culture.

• Journal of Aerospace System Engineering
• /
• v.12 no.5
• /
• pp.40-47
• /
• 2018

The objective of this study is to investigate the turbulence models with compressibility correction for large separation-flow in a supersonic convergent-divergent rectangular nozzle. As turbulence models, Yang and Shih's Low-Re $k-{\varepsilon}$ model, Mener's $k-{\omega}$ SST model and Wilcox's $k-{\omega}$model were evaluated. In order to get a significant compressible effects, Sarkar and Wilcox compressibility correction models were applied to the turbulence models respectively. Also, the simulation results were compared with experimental data. The turbulence model with compressibility correction model improves both of shock position and pressure recovery, but deteriorates the length of Mach disk.

• Korean Journal of Soil Science and Fertilizer
• /
• v.34 no.1
• /
• pp.64-70
• /
• 2001

To verify the newly developed decision assisting expert system for nitrogen fertilizer application NES to Illinois cornfields, a couple of N rate studies from Dr. Howard and five Illinois Agricultural Experiment Stations were applied. Four types of recommendations including the current Illinois recommendation, Hoeft recommendation, NES, and maximum economic recommendation were compared with each other for the crop yields, profits, recovery rate, and N losses to cornfields. The N rate of NES recommendation, considering productivity index (PI), soil organic matter content (SOM), and pre-sidedressing nitrate concentration (PSNT) level, was the lowest in comparison to those of other recommendations. However, N recovery rate in NES was generally higher and the resulting N loss was lower than others. But, adherence to the recommendations may also reduce farmers income if environmental expense did not considered. Therefore, NES will be more effective by adding the factors including environmental expense, tillage systems, crop rotation, and other agricultural management parameters.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2017.10a
• /
• pp.454-456
• /
• 2017

Current and capacitance-voltage characteristics of tunnel field effect transistor (TFET) with various quantum models were investigated. Density gradient, Bohm quantum potential (BQP), and Vandort quantum correction are used with calibrating against Schrodinger-Poisson model. Drive-currents in all models. are decreased. When only BQP is used, SS and $V_{onset}$ are fixed but drive-current is decreased 3 times more than those of no quantum model. And When BQP with Vandort and density gradient are used, SS increased more than 40 mV./dec and $V_{onset}$ shifted as 0.07 eV.

• Annual Conference on Human and Language Technology
• /
• 2022.10a
• /
• pp.59-64
• /
• 2022

한국표준질병사인분류(KCD)는 사람의 질병과 사망 원인을 유사성에 따라 체계적으로 유형화한 분류체계이다. KCD는 계층적 분류체계로 구성되어 있어 분류마다 연관성이 존재하지만, 일반적인 텍스트 분류 모델은 각각의 분류를 독립적으로 예측하기 때문에 계층적 정보를 반영하는 데 한계가 있다. 본 논문은 계층적 분류체계를 적용한 KCD 예측 모델을 제안한다. 제안 방법의 효과를 입증하기 위해 비교 실험을 진행한 결과 F1-score 기준 최대 0.5%p의 성능 향상을 확인할 수 있었다. 특히 비교 모델이 잘 예측하지 못했던 저빈도의 KCD에 대해서 제안 모델은 F1-score 기준 최대 1.1%p의 성능이 향상되었다.

• The KIPS Transactions:PartD
• /
• v.11D no.1
• /
• pp.105-114
• /
• 2004

In this paper, we propose a collaboration method of diverse agents each others in multi-agent model and describe a node migration algorithm of Mobile-Agent (MA) using by the metadata of Naming-Agent (NA). Collaboration work of multi-agent assures stability of agent system and provides reliability of information retrieval on the distributed environment. NA, an important part of multi-agent, identifies each agents and series the unique name of each agents, and each agent references the specified object using by its name. Also, NA integrates and manages naming service by agents classification such as Client-Push-Agent (CPA), Server-Push-Agent (SPA), and System-Monitoring-Agent (SMA) based on its characteristic. And, NA provides the location list of mobile nodes to specified MA. Therefore, when MA does move through the nodes, it is needed to improve the efficiency of node migration by specified priority according to hit_count, hit_ratio, node processing and network traffic time. Therefore, in this paper, for the integrated naming service, we design Naming Agent and show the structure of metadata which constructed with fields such as hit_count, hit_ratio, total_count of documents, and so on. And, this paper presents the flow of creation and updating of metadata and the method of node migration with hit_count through the collaboration of multi-agent.

• Proceedings of the Korea Concrete Institute Conference
• /
• 2008.04a
• /
• pp.277-280
• /
• 2008

In this paper behavior of U-Channel Bridge (UCB) was studied, and a new analysis model was proposed. Most of the time, permanent and traffic load actions are directly transmitted to main beams located under the carriageway, one of the most distinctive features of UCB is that the edge beams that support the bridge are above the deck, in contrast with a conventional overpass system. In This study models used with the frame elements, the frame and plate elements, and the solid elements were constructed. Assuming that the results of solid models were similar to the real behavior of UCB, results of another models was compared. The results of the models used with the frame and plate elements were similar to the results of solid models, the model used with the frame and plate elements was proposed as an analysis model.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.24 no.5
• /
• pp.584-590
• /
• 2020

Long-term use of digital screens in daily life can lead to computer vision syndrome including symptoms such as eye strain, dry eyes, and headaches. To prevent computer vision syndrome, it is important to limit screen usage time and take frequent breaks. There are a variety of applications that can help users know the screen usage time. However, these apps are limited because users see various screens such as desktops, laptops, and tablets as well as smartphone screens. In this paper, we propose and evaluate machine learning-based models that detect the screen device in use using color, IMU and lidar sensor data. Our evaluation shows that neural network-based models show relatively high F1 scores compared to traditional machine learning models. Among neural network-based models, the MLP and CNN-based models have higher scores than the LSTM-based model. The RF model shows the best result among the traditional machine learning models, followed by the SVM model.