• Proceedings of the Korea Information Processing Society Conference
• /
• 2024.05a
• /
• pp.602-603
• /
• 2024

이 논문은 코로나 팬데믹 시대에 원격 의료 서비스의 중요성이 부상함에 따라, LLM(Large Language Model)과 웨어러블 기기를 활용한 의료 기술의 발전과 이를 통한 의료 서비스의 혁신에 대해 다루고 있다. 코로나 19 대응을 위해 원격 의료에 대한 법적 제한이 완화되며, 이에 따른 원격 의료 시스템의 확대를 언급하고 있다. LLM 을 활용한 의료 정보 관리와 웨어러블을 통한 건강 모니터링을 소개하며, 대화형 AI 를 통한 문의사항 처리와 2 차 처방, 실시간 번역 AI 기술 등의 기술적 혁신을 언급하고 있다. 이러한 기술들이 의료 서비스의 혁신과 개인 건강 관리에 새로운 차원을 열어주지만, 보안 문제와 디지털 격차 등의 문제가 동반될 수 있다고 경고하며, 이를 극복하기 위한 대책과 지속적인 개선이 필요하다고 강조하고 있다.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2024.05a
• /
• pp.786-787
• /
• 2024

Large Language Model은 그 기능으로 말미암아 여러 애플리케이션에 통합되고 있다. 특히 OpenAI는 ChatGPT에 여러 세부 사항을 설정함으로써 차별화된 기능을 사용자가 제공할 수 있도록 한다. 하지만 최근 제시되는 프롬프트 연출 공격은 서비스의 핵심 요소를 쉽게 탈취할 수 있는 가능성을 제시한다. 본 연구는 지침 우회 방법론을 통해 기본 대비 공격의 성공률을 10%p 올렸다. 또한 유출공격을 평가할 수 있는 유효성과 성공률을 통해 모델의 방어 성능을 일반화한다.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2024.05a
• /
• pp.733-736
• /
• 2024

최근 GTP4, LLama와 같은 초거대 언어모델을 활용한 서비스가 공개되어 많은 사람의 주목을 받고 있다. 해당 모델들은 사용자들의 다양한 질문에 대하여 유창한 결과를 생성하고 있지만 한국어 데이터에 대한 학습량이 부족하여 한국어 이해 및 한국 문화 등에 대한 잘못된 정보를 제공하는 문제를 야기할 수 있다. 이에 본 논문에서는 한국어 데이터를 학습한 주요 공개 모델 6개를 선정하고 5개 분야(한국어 이해 및 문화 영역으로 구성)에 대한 평가 데이터셋을 구성하여 한국어 이해 능력에 대한 평가를 진행하였다. 그 결과 한국어 구사 능력은 Bookworm 모델이, 한국어 이해 및 문화와 관련한 부문은 LDCC-SOLAR 모델이 우수한 것으로 확인할 수 있었다.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2024.05a
• /
• pp.745-748
• /
• 2024

글로벌 시장에서 Large Language Model(LLM)의 발전이 급속하게 이루어지며 활용도가 높아지고 있지만 특정 유형이나 전문적 지식이 부족할 수 있어 일반화하기 어려우며, 새로운 데이터로 업데이트하기 어렵다는 한계점이 있다. 이를 극복하기 위해 지속적으로 업데이트되는 최신 정보를 포함한 외부 데이터베이스에서 정보를 검색해 응답을 생성하는 Retrieval- Augmented Generation(RAG, 검색 증강 생성) 모델을 도입하여 LLM의 환각 현상을 최소화하고 효율성과 정확성을 향상시키려는 연구가 활발히 이루어지고 있다. 본 논문에서는 LLM의 검색 기능을 강화하기 위한 RAG의 연구 및 평가기법에 대한 최신 연구 동향을 소개하고 실제 산업에서 활용하기 위한 최적화 및 응용 사례를 소개하며 이를 바탕으로 향후 연구 방향성을 제시하고자 한다.

• Journal of Environmental Science International
• /
• v.7 no.6
• /
• pp.745-751
• /
• 1998

The well-mixed room model has been traditionally used to predict the concentrations of contaminants in indoor environments. However, this is inappropriate because the flow fields in many indoor environments distribute contaminants non-uniformly, due to imperfect air mixing. Thus, some means used to describe an imperfectly mixed room are needed. The simplest model that accounts for imperfect air mixing is a two-zone model. Therefore, this study on development of computer program far the two-zone model is carried out to propose techniques of estimating the concentration of contaminants in the room. To do this, an important consideration is to divide a room into two-zone, i.e. the lower and upper zone assuming that the air and contaminants are well mixed within each zone. And between the zones the air recirculation is characterized through the air exchange parameter. By this basic assumption, the equations for the conservation of mass are derived for each zone. These equations are solved by using the computational technique. The language used to develope the program is a VISUAL BASIC. The value of air exchange coefficient($f_12$) is the most difficult to forecast when the concentrations of contaminants in an imperfectly mixed room are estimated by the two-zone model. But, as the value of $f_12$ increases, the air exchange between each zone increases. When the value of $f_12$ is approximately 15, the concentrations in both zone approach each other, and the entire room may be approximately treated as a single well-mixed room. Therefore, this study is available for designing of the ventilation to improve the air quality of indoor environments. Also, the two-zone model produces the theoretical base which may be extended to the theory for the multi-zone model, that will be contributed to estimate the air pollution in large enclosures, such as shopping malls, atria buildings, atria terminals, and covered sports stadia.

• Journal of Korea Society of Industrial Information Systems
• /
• v.29 no.3
• /
• pp.1-12
• /
• 2024

In this paper, we fine-tuned KoAlpaca, a large-scale Korean language model, and implemented a blog text generation system utilizing it. Blogs on social media platforms are widely used as a marketing tool for businesses. We constructed training data of positive reviews through emotion analysis and refinement of collected sponsored blog texts and applied QLoRA for the lightweight training of KoAlpaca. QLoRA is a fine-tuning approach that significantly reduces the memory usage required for training, with experiments in an environment with a parameter size of 12.8B showing up to a 58.8% decrease in memory usage compared to LoRA. To evaluate the generative performance of the fine-tuned model, texts generated from 100 inputs not included in the training data produced on average more than twice the number of words compared to the pre-trained model, with texts of positive sentiment also appearing more than twice as often. In a survey conducted for qualitative evaluation of generative performance, responses indicated that the fine-tuned model's generated outputs were more relevant to the given topics on average 77.5% of the time. This demonstrates that the positive review generation language model for sponsored content in this paper can enhance the efficiency of time management for content creation and ensure consistent marketing effects. However, to reduce the generation of content that deviates from the category of positive reviews due to elements of the pre-trained model, we plan to proceed with fine-tuning using the augmentation of training data.

• Computers and Concrete
• /
• v.20 no.3
• /
• pp.319-327
• /
• 2017

Studying the structural behavior of prestressed segmented girders is quite important due to the large use this type of solution in viaducts and bridges. Thus, this work presents a nonlinear three-dimensional structural analysis of an externally prestressed segmented concrete girder through the Finite Element Method (FEM), using a customized ANSYS platform, version 14.5. Aiming the minimization of the computational effort by using the lowest number of finite elements, a new viscoelastoplastic material model has been implemented for the structural concrete with the UPF customization tool of ANSYS, adding new subroutines, written in FORTRAN programming language, to the main program. This model takes into consideration the cracking of concrete in its formulation, being based on fib Model Code 2010, which uses Ottosen rupture surface as the rupture criterion. By implementing this new material model, it was possible to use the three-dimensional 20-node quadratic element SOLID186 to model the concrete. Upon validation of the model, an externally prestressed segmented box concrete girder that was originally lab tested by Aparicio et al. (2002) has been computationally simulated. In the discretization of the structure, in addition to element SOLID186 for the concrete, unidimensional element LINK180 has been used to model the prestressing tendons, as well as contact elements CONTA174 and TARGE170 to simulate the dry joints along the segmented girder. Stresses in the concrete and in the prestressing tendons are assessed, as well as joint openings and load versus deflection diagrams. A comparison between numerical and experimental data is also presented, showing a good agreement.

• The Transactions of the Korea Information Processing Society
• /
• v.7 no.10
• /
• pp.3075-3085
• /
• 2000

According to distributed, large-caled environment of software development and operation, the elicitation and specivication of correct and complete requrement is the most important factor for the system. In addition contiuous and dramatic systerm canging requests in cooperative environment with internet require more efficient, requirement management. In this paper we detine the specification architecture and techruques for requrements, so that we improve the efficiency ofnatural language-based requirement speciticationand management in a cooperatie work environment. Also, we propose a software requirement engineering process model and environment based on requirements in a CSC@(Computer Supported Cjooperative Work) environment, therefore transfer them into analysis phase.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.16 no.12
• /
• pp.3868-3888
• /
• 2022

A widely used social networking service like Twitter has the ability to disseminate information to large groups of people even during a pandemic. At the same time, it is a convenient medium to share irrelevant and unverified information online and poses a potential threat to society. In this research, conventional machine learning algorithms are analyzed to classify the data as either non-rumor data or rumor data. Machine learning techniques have limited tuning capability and make decisions based on their learning. To tackle this problem the authors propose a deep learning-based Rumor Detection Neural Network model to predict the rumor tweet in real-world events. This model comprises three layers, AttCNN layer is used to extract local and position invariant features from the data, AttBi-LSTM layer to extract important semantic or contextual information and HPOOL to combine the down sampling patches of the input feature maps from the average and maximum pooling layers. A dataset from Kaggle and ground dataset #gaja are used to train the proposed Rumor Detection Neural Network to determine the veracity of the rumor. The experimental results of the RDNN Classifier demonstrate an accuracy of 93.24% and 95.41% in identifying rumor tweets in real-time events.

• Journal of Engineering Education Research
• /
• v.26 no.3
• /
• pp.12-19
• /
• 2023

The purpose of this study is to develop the Artificial Intelligence thinking-based education program for improving AI literacy and verify its effectiveness for beginner. This program consists of 17 sessions, was designed according to the "ABCDE" model and is a project-based program. This program was conducted on 51 first-year middle school students and 36 respondents excluding missing values were analyzed in R language. The effect of this program on ethics, understanding, social competency, execution plan, data literacy, and problem solving of AI literacy is statistically significant and has very large practical significance. According to the result of this study, this program provided learners experiencing Artificial Intelligence education for the first time with Artificial Intelligence concepts and principles, collection and analysis of information, and problem-solving processes through application in real life, and served as an opportunity to enhance AI literacy. In addition, education program to enhance AI literacy should be designed based on AI thinking.