• ETRI Journal
• /
• 제41권6호
• /
• pp.771-781
• /
• 2019

We propose the VS³-NET model to solve the task of question answering questions with machine-reading comprehension that searches for an appropriate answer in a given context. VS³-NET is a model that trains latent variables for each question using variational inferences based on a model of a simple recurrent unit-based sentences and self-matching networks. The types of questions vary, and the answers depend on the type of question. To perform efficient inference and learning, we introduce neural question-type models to approximate the prior and posterior distributions of the latent variables, and we use these approximated distributions to optimize a reparameterized variational lower bound. The context given in machine-reading comprehension usually comprises several sentences, leading to performance degradation caused by context length. Therefore, we model a hierarchical structure using sentence encoding, in which as the context becomes longer, the performance degrades. Experimental results show that the proposed VS³-NET model has an exact-match score of 76.8% and an F1 score of 84.5% on the SQuAD test set.

• 정보처리학회논문지:소프트웨어 및 데이터공학
• /
• 제11권7호
• /
• pp.307-314
• /
• 2022

최근 전력수요를 예측하기 위해 통계기반 시계열 분석 기법을 대체하기 위해 딥러닝 기법을 활용한 연구가 활발히 진행되고 있다. 딥러닝 기반 전력수요 예측 연구 결과를 분석한 결과, LSTM 기반 예측 모델의 성능이 우수한 것으로 규명되었으나 장기간의 지역 범위 전력수요 예측에 대해 LSTM 기반 모델의 성능이 충분하지 않음을 확인할 수 있다. 본 연구에서는 기온 데이터를 반영하여 24시간 이전에 전력수요를 예측하는 WaveNet 기반 딥러닝 모델을 개발하여, 실제 사용하고 있는 통계적 시계열 예측 기법의 정확도(MAPE 값 2%)보다 우수한 예측 성능을 달성하는 모델을 개발하고자 한다. 먼저 WaveNet의 핵심 구조인 팽창인과 1차원 합성곱 신경망 구조를 소개하고, 전력수요와 기온 데이터를 입력값으로 모델에 주입하기 위한 데이터 전처리 과정을 제시한다. 다음으로, 개선된 WaveNet 모델을 학습하고 검증하는 방법을 제시한다. 성능 비교 결과, WaveNet 기반 모델에 기온 데이터를 반영한 방법은 전체 검증데이터에 대해 MAPE 값 1.33%를 달성하였고, 동일한 구조의 모델에서 기온 데이터를 반영하지 않는 것(MAPE 값 2.31%)보다 우수한 전력수요 예측 결과를 나타내고 있음을 확인할 수 있다.

• 한국게임학회 논문지
• /
• 제15권5호
• /
• pp.49-56
• /
• 2015

캐주얼게임이 대중화되면서 게임개발과정에서 다양한 유저들의 게임플레이의 성향과 요구사항들을 만족시켜야 할 필요성이 증가하였다. 이를 위해서는 게임개발과정의 테스트단계에서 다양한 유저의 게임플레이 패턴을 분석해야 한다. 본 논문에서는 유저의 게임플레이 측정데이터를 사용하는 Petri net 모델의 시뮬레이션을 통해 액션 패턴을 분석하는 방법을 제안하였다. 제안한 방법은 유저의 게임플레이 측정데이터를 사용하기 때문에 시뮬레이션 환경은 실제적이고 또한 Petri net 모델을 사용한 분석이기 때문에 액션 패턴의 reachability, coverbility, liveness 등과 같은 다양한 분석이 가능하다. 제안하는 방법의 적용사례로 Petri net 모델링 도구인 GPenSIM v4.0 도구를 사용하여 팩맨(Pacman) 게임의 게임플레이 패턴을 분석하는 Petri net 모델을 구현하고 시뮬레이션을 결과들을 제시하였다. 적용사례의 제시 결과들은 제안하는 방법이 Petri net 분석 기능을 이용하여 유저의 게임플레이의 액션패턴을 다양하게 분석가능 함을 보여주었다.

• 스마트미디어저널
• /
• 제9권3호
• /
• pp.59-70
• /
• 2020

This paper presents a fully automatic tool to recognize the liver region from CT images based on a deep learning model, namely Multiple Filter U-net, MFUnet. The advantages of both U-net and Multiple Filters were utilized to construct an autoencoder model, called MFUnet for segmenting the liver region from computed tomograph. The MFUnet architecture includes the autoencoding model which is used for regenerating the liver region, the backbone model for extracting features which is trained on ImageNet, and the predicting model used for liver segmentation. The LiTS dataset and Chaos dataset were used for the evaluation of our research. This result shows that the integration of Multiple Filter to U-net improves the performance of liver segmentation and it opens up many research directions in medical imaging processing field.

• 한국컴퓨터정보학회논문지
• /
• 제21권12호
• /
• pp.115-124
• /
• 2016

In this paper, a method to generate the extended model automatically on the critical elements in enterprise system based real time distributed architecture as well as the platform specific model(PSM) for Microsoft(MS) .NET platform is proposed. The key ideas of this method are real time distributed architecture should performed with satisfying strict constraints on life cycle of object and response time such as synchronization, transaction and so on, and .NET platform is able to implement functionalities including before mentioned by only specifying Attribute Code and maximizing advantages of MDA. In order to realize the ideas, functionalities which should be considered enterprise system development are specified and these are to be defined in Meta Model and extended UML profile. In addition, after definition of UML profile for .NET specification, by developing and applying these into plug-in of open source MDA tool, and extended models are generated automatically through this tool. Accordingly, by using proposed specification technology, the profile and tools easily and quickly reusable extended model can be generated even though low level of detailed information for functionalities which is considered in .NET platform and real time distributed architecture. In addition, because proposed profile is MOF which is basis of standard extended and applied, UML and MDA tools which observed MOF is reusable.

• 대한건축학회논문집:구조계
• /
• 제35권4호
• /
• pp.185-192
• /
• 2019

In this study, we, through case studies, formulated a method to implement net-zero energy daycare center at the current insulation and technology level, and calculated its energy expense. The reference model was a medium sized daycare center whose number of children was 99. We analyzed the energy consumption status for the reference model and developed TRNSYS simulation analytical model to realize net-zero energy . We assumed the reference model to be "All Electric Building" where all energy including cooking is supplied by electricity. The result is summarized as follows: First, the annual electricity consumption of daycare center was 53,291kWh. Plug load occupied the largest share of 48% followed by lighting, 10%, cooling, 9%, cooking, 9%, heating, 8%, hot water, 5% and ventilation, 2%. Second, the photovoltaic installation capacity to realize net-zero energy was 40.32kWp and its annual generation was 53,402kWh. Third, the annual energy expense(electricity bill) by realizing net-zero energy was 2,620,390won.

• Journal of the Korean Wood Science and Technology
• /
• 제47권3호
• /
• pp.265-276
• /
• 2019

In our previous study, the LeNet3 model successfully classified images from the transverse surfaces of five Korean softwood species (cedar, cypress, Korean pine, Korean red pine, and larch). However, a practical limitation exists in our system stemming from the nature of the training images obtained from the transverse plane of the wood species. In real-world applications, it is necessary to utilize images from the longitudinal surfaces of lumber. Thus, we improved our model by training it with images from the longitudinal and transverse surfaces of lumber. Because the longitudinal surface has complex but less distinguishable features than the transverse surface, the classification performance of the LeNet3 model decreases when we include images from the longitudinal surfaces of the five Korean softwood species. To remedy this situation, we adopt ensemble methods that can enhance the classification performance. Herein, we investigated the use of ensemble models from the LeNet and MiniVGGNet models to automatically classify the transverse and longitudinal surfaces of the five Korean softwoods. Experimentally, the best classification performance was achieved via an ensemble model comprising the LeNet2, LeNet3, and MiniVGGNet4 models trained using input images of $128{\times}128{\times}3pixels$ via the averaging method. The ensemble model showed an F1 score greater than 0.98. The classification performance for the longitudinal surfaces of Korean pine and Korean red pine was significantly improved by the ensemble model compared to individual convolutional neural network models such as LeNet3.

• Journal of Information Processing Systems
• /
• 제15권6호
• /
• pp.1472-1488
• /
• 2019

Critical elements in the China's Lunar Exploration reside in that the lunar rover travels over the surrounding undetermined environment and it conducts scientific exploration under the ground control via teleoperation system. Such an interplanetary transportation mission teleoperation system belongs to the ground application system in deep space mission, which performs terrain reconstruction, visual positioning, path planning, and rover motion control by receiving telemetry data. It plays a vital role in the whole lunar exploration operation and its so-called trustworthy evidence must be assessed before and during its implementation. Taking ISO standards and China's national military standards as trustworthy evidence source, the net assessment model and net assessment method of teleoperation system are established in this paper. The multi-dimensional net assessment model covering the life cycle of software is defined by extracting the trustworthy evidences from trustworthy evidence source. The qualitative decisions are converted to quantitative weights through the net assessment method (NAM) combined with fuzzy analytic hierarchy process (FAHP) and entropy weight method (EWM) to determine the weight of the evidence elements in the net assessment model. The paper employs the teleoperation system for interplanetary transportation as a case study. The experimental result drawn shows the validity and rationality of net assessment model and method. In the final part of this paper, the untrustworthy elements of the teleoperation system are discovered and an improvement scheme is established upon the "net result". The work completed in this paper has been applied in the development of the teleoperation system of China's Chang'e-3 (CE-3) "Jade Rabbit-1" and Chang'e-4 (CE-4) "Jade Rabbit-2" rover successfully. Besides, it will be implemented in China's Chang'e-5 (CE-5) mission in 2019. What's more, it will be promoted in the Mars exploration mission in 2020. Therefore it is valuable to the development process improvement of aerospace information system.

• 대한원격탐사학회지
• /
• 제37권5_1호
• /
• pp.1061-1070
• /
• 2021

본 연구에서는 국내 농업지역에 대한 작물재배지역의 분류를 위하여 FC-DenseNet 모델에 attention gate를 적용하여 딥러닝 모델의 성능을 향상시키고자 하였다. Attention gate는 특징맵의 공간/분광적 중요도에 따른 가중치를 추가적으로 학습하여 딥러닝 모델의 학습을 용이하게 하고, 모델의 성능을 향상시킬 수 있다. Attention gate를 FC-DenseNet의 스킵 연결 부분에 추가한 딥러닝 모델을 이용하여 양파 및 마늘 지역의 작물분류를 수행하였다. PlanetScope 위성영상을 이용하여 훈련자료를 제작하였으며, 훈련자료의 불균형 문제를 해결하기 위하여 전처리 과정을 적용하였다. 다양한 평가자료를 이용하여 작물재배분류 결과를 평가한 결과, 제안된 딥러닝 모델은 기존의 FC-DenseNet과 비교하여 효과적으로 양파 및 마늘 지역을 분류할 수 있는 것을 확인하였다.

• The Korean Journal of Ecology
• /
• 제9권1호
• /
• pp.41-50
• /
• 1986

The net primary production in the Korean peninsula was estimated by Miami model, Montreal model and Kira's model, based on 148 meteorological data. The modes in frequency distribution of the values calculated by Montreal and Miami model were found at 1,500g/m2/yr. class and at one step high class in 100g. interval, while by Kira's madel at 1,700g/m2/yr. class. The relationships between values by Miami model(X) and those by Motreal model (Ym) and Kira's model(Yk) can be expressed as follows: Ym=0.365X+944.7, Yk=0.462 X+1006.9 and Yk=1.282Ym-211.5. The total amount of the net primary production in 218,583.4km2, 98.9% of the whole area(220,951 km2) of the Korean Peninsula, was estimated as 290,691,407 tons/yr. by Miami model, 310,751,566 tons/yr by Montreal model and 352,071,901 tons/yr by Kira's model. Therefore, it is reasonable that the organic substance over 300 million-tons is added yearly in the Korean Peninsula, because only 1.1% of the whole area no calculated. In additiion, the net primary production amount of Han-river basin was estimated as ca. 38 million-tons, whether calculated with the meteorological data in level of the Korean Peninsula or with more detail data.