• Annual Conference on Human and Language Technology
• /
• 2020.10a
• /
• pp.9-13
• /
• 2020

관계 추출은 주어진 문장이나 문서에 존재하는 개체들 간의 의미적 관계를 찾아내는 작업을 말한다. 최근 문서 수준 관계 추출 말뭉치인 DocRED가 공개되면서 문서 수준 관계 추출에 대한 연구가 활발히 진행되고 있다. 또한 사전 학습된 Masked Language Model(MLM)이 자연어처리 분야 전체에 영향력을 보이면서 관계 추출에서도 MLM을 사용하는 연구가 진행되고 있다. 그러나 문서 수준의 관계 추출은 문서의 단위가 길기 때문에 Self-attention을 기반으로 하는 MLM을 사용하면 모델의 계산량이 증가하는 문제가 있다. 본 논문은 이 점을 보완하기 위해 관계 추출에 필요한 문장을 선별하는 간단한 전처리 방법을 제안한다. 또한 문서의 길이에 상관없이 관계 추출에 필요한 어휘 정보를 자동으로 습득 할 수 있는 Relation-Context Co-attention 방법을 제안한다. 제안 모델은 DocRED 말뭉치에서 Dev F1 62.01%, Test F1 59.90%로 높은 성능을 보였다.

• The Korean Journal of Applied Statistics
• /
• v.34 no.2
• /
• pp.191-204
• /
• 2021

Visual Question Answering (VQA) and image captioning are tasks that require understanding of the features of images and linguistic features of text. Therefore, co-attention may be the key to both tasks, which can connect image and text. In this paper, we propose a model to achieve high performance for VQA by image caption generated using a pretrained standard transformer model based on MSCOCO dataset. Captions unrelated to the question can rather interfere with answering, so some captions similar to the question were selected to use based on a similarity to the question. In addition, stopwords in the caption could not affect or interfere with answering, so the experiment was conducted after removing stopwords. Experiments were conducted on VQA-v2 data to compare the proposed model with the deep modular co-attention network (MCAN) model, which showed good performance by using co-attention between images and text. As a result, the proposed model outperformed the MCAN model.

• Proceedings of the Korean Society of Computer Information Conference
• /
• 2022.07a
• /
• pp.331-332
• /
• 2022

본 논문에서는 3축 가속도와 3축 각속도 센서로 구성된 관성 측정 장치(IMU)와 압력센서가 내장되어있는 스마트 인솔을 착용하여 얻어진 보행 데이터를 통해 보폭을 추정하는 방법을 제안한다. 먼저 압력센서를 활용하여 한 걸음 주기로 나눈 뒤 나누어진 가속도와 각속도 센서 데이터를 LSTM과 Attention 계층을 결합한 딥러닝 모델에 학습하여 보폭 추정을 시행하였다. LSTM-Attention 모델은 기존 LSTM 모델보다 약 1.14%의 성능 향상을 보였다.

• Annual Conference on Human and Language Technology
• /
• 2020.10a
• /
• pp.37-41
• /
• 2020

본 논문에서는 Joint CTC/Attention 모델에 CTC ratio scheduling을 이용한 end-to-end 한국어 음성인식을 연구하였다. Joint CTC/Attention은 CTC와 attention의 장점을 결합한 모델로서 attention, CTC 단일 모델보다 좋은 성능을 보여주지만, 학습이 진행될수록 CTC가 attention의 학습을 저해하는 요인이 된다. 본 논문에서는 이러한 문제를 해결하기 위해, 학습 진행에 따라 CTC의 비율(ratio)를 줄여나가는 CTC ratio scheduling 방법을 제안한다. CTC ratio scheduling를 이용하여 학습한 결과물은 기존 Joint CTC/Attention, 단일 attention 모델 대비 좋은 성능을 보여주는 것을 확인하였다.

• Journal of Biomedical Engineering Research
• /
• v.41 no.5
• /
• pp.195-202
• /
• 2020

Needle detection in ultrasound images is sometimes difficult due to obstruction of fat tissues. Accurate needle detection using continuous ultrasound (CUS) images is a vital stage of treatment planning for tissue biopsy and brachytherapy. The main goal of the study is classified into two categories. First, new detection model, i.e. D-Attention Unet, is developed by combining the context information of 3D medical data and CUS images. Second, the D-Attention Unet model was compared with other models to verify its usefulness for needle detection in continuous ultrasound images. The continuous needle images taken with ultrasonic waves were converted into still images for dataset to evaluate the performance of the D-Attention Unet. The dataset was used for training and testing. Based on the results, the proposed D-Attention Unet model showed the better performance than other 3 models (Unet, D-Unet and Attention Unet), with Dice Similarity Coefficient (DSC), Recall and Precision at 71.9%, 70.6% and 73.7%, respectively. In conclusion, the D-Attention Unet model provides accurate needle detection for US-guided biopsy or brachytherapy, facilitating the clinical workflow. Especially, this kind of research is enthusiastically being performed on how to add image processing techniques to learning techniques. Thus, the proposed method is applied in this manner, it will be more effective technique than before.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.16 no.2
• /
• pp.544-564
• /
• 2022

Underwater images often suffer from color distortion, blurring and low contrast, which is caused by the propagation of light in the underwater environment being affected by the two processes: absorption and scattering. To cope with the poor quality of underwater images, this paper proposes a multiscale fusion underwater image enhancement method based on channel attention mechanism and local binary pattern (LBP). The network consists of three modules: feature aggregation, image reconstruction and LBP enhancement. The feature aggregation module aggregates feature information at different scales of the image, and the image reconstruction module restores the output features to high-quality underwater images. The network also introduces channel attention mechanism to make the network pay more attention to the channels containing important information. The detail information is protected by real-time superposition with feature information. Experimental results demonstrate that the method in this paper produces results with correct colors and complete details, and outperforms existing methods in quantitative metrics.

• Journal of the Korean Association of Geographic Information Studies
• /
• v.26 no.4
• /
• pp.116-129
• /
• 2023

Recently, interest in autonomous driving has drawn attention to autonomous cooperative driving, which considers the development of driving technology of autonomous vehicles and the development of infrastructure that constitutes a driving environment. According to the concept of autonomous cooperative driving, This study analyzes the new infrastructure for autonomous driving that can complement the information of existing precise road maps and adding HD map layer as the new infrastructure. The new infrastructure for autonomous driving presented two types of improved facilities and one type of sensor only facility. Analysis of HD maps shows that information such as junction points rarely changes, but it is expected that infrastructure for autonomous driving can be added to convey the meaning of paying attention to obstacles that may arise at the junction. In this way, the new infrastructure for autonomous driving needs to support the roles of guidance, instruction, and attention that existing road facilities.

• Proceedings of the KSME Conference
• /
• 2007.05b
• /
• pp.2648-2653
• /
• 2007

Recently, increasing demand on not only lighter but also extremely mobile battery make micro fuel cell device very attractive alternative. By reducing the size of fuel cell, surface tension becomes dominant factor with minor gravitational effect. Therefore, it is very difficult to detach the $CO_2$ bubble generating on a cathode side in ${\mu}DMFC$ (micro direct methanol fuel cell). The degassing of a $CO_2$ bubble has drawn quite attention especially for ${\mu}DMFC$ due to its considerable effect on overall machine performance. Our attention has been paid to the dynamic behavior of immiscible bubble attached to the one side of the wall on 2D rectangular channel subject to external shear flow. We use Level Contour Reconstruction Method (LCRM) which is simplified version of front tracking method to track the bubble interface motion. Effects of Reynolds number, Weber number, advancing/receding contact angle and property ratio on bubble detachment characteristic has been numerically identified.

• Journal of Information Technology Services
• /
• v.15 no.3
• /
• pp.157-171
• /
• 2016

As unemployment rates and concerns about turnover keep growing, the need for information is also increasing. In these situations, the job reviews which share information about the company catch people's attention because they are usually created by people who worked at the company. The development of SNS and mobile environments has led to an increase in the web services that provide job reviews. For example, Jobplanet is a job review service in Korea, and Glassdoor.com offers a similar service in the US. Despite this attention, however, research utilizing job reviews is insufficient. This paper asks whether there are differences in ratios of job satisfaction factors by industry, using LDA topic modeling and co-occurrence analysis to explore the differences. Through the results of LDA, we find that the ratios of job satisfaction factors are similar by industry. At the same time, the results of co-occurrence analysis show that the co-occurrence frequency of some job satisfaction factors appears high: pay and welfare, balance of work and life, company culture. We expect that the result of this research will be helpful in comparative analysis of job satisfaction factors by industry. Furthermore, in this paper we suggest how to use the job review data in organizational behavior research.

• Journal of Climate Change Research
• /
• v.7 no.2
• /
• pp.177-184
• /
• 2016

Recently, the need for technology development of commercial vehicle fuel consumption has emerged. Fuel economy improvement of transport equipment and transportation efficiency, and increasing attention to the logistics cost reduction measures. Increasing attention to the logistics cost reduction measures by fuel economy improvement of transport equipment and transportation efficiency. In this study, we have installed aerodynamic reduction device (side skirt, boat tail) to 14.5 ton cargo trucks and 45 ft tractor-trailers. And the fuel consumption was compared installed before and after. Fuel economy assessment for the aerodynamic reduction value device was tested by modifying the SAE J1321 Joint TMC/SAE Fuel Consumption Test Procedure - Type II test in according domestic situation. Greenhouse gas reductions were calculated in accordance with the scenario, including fuel consumption test results. When the 14.5 ton cargo trucks has been equipped with side skirts and boat tail, it confirmed the improvement in fuel efficiency of 4.72%. One Heavy-duty truck's the annual greenhouse gas reductions value are $6.86ton\;CO_2\;eq$. And if applying the technology to more than 50% of registered 15 ton trucks, greenhouse gas reductions are calculated as $686,826ton\;CO_2\;eq./yr$.