• The Sea:JOURNAL OF THE KOREAN SOCIETY OF OCEANOGRAPHY
• /
• v.28 no.1
• /
• pp.1-18
• /
• 2023

The East Sea, one of the regions where the most rapid warming is occurring, is known to have important implications for the response of the ocean to future climate changes because it not only reacts sensitively to climate change but also has a much shorter turnover time (hundreds of years) than the ocean (thousands of years). However, the processes underlying changes in seawater characteristics at the sea's deep and abyssal layers, and meridional overturning circulation have recently been examined only after international cooperative observation programs for the entire sea allowed in-situ data in a necessary resolution and accuracy along with recent improvement in numerical modeling. In this review, previous studies on the physical characteristics of seawater at deeper parts of the East Sea, and meridional overturning circulation are summarized to identify any remaining issues. The seawater below a depth of several hundreds of meters in the East Sea has been identified as the Japan Sea Proper Water (East Sea Proper Water) due to its homogeneous physical properties of a water temperature below 1℃ and practical salinity values ranging from 34.0 to 34.1. However, vertically high-resolution salinity and dissolved oxygen observations since the 1990s enabled us to separate the water into at least three different water masses (central water, CW; deep water, DW; bottom water, BW). Recent studies have shown that the physical characteristics and boundaries between the three water masses are not constant over time, but have significantly varied over the last few decades in association with time-varying water formation processes, such as convection processes (deep slope convection and open-ocean deep convection) that are linked to the re-circulation of the Tsushima Warm Current, ocean-atmosphere heat and freshwater exchanges, and sea-ice formation in the northern part of the East Sea. The CW, DW, and BW were found to be transported horizontally from the Japan Basin to the Ulleung Basin, from the Ulleung Basin to the Yamato Basin, and from the Yamato Basin to the Japan Basin, respectively, rotating counterclockwise with a shallow depth on the right of its path (consistent with the bottom topographic control of fluid in a rotating Earth). This horizontal deep circulation is a part of the sea's meridional overturning circulation that has undergone changes in the path and intensity. Yet, the linkages between upper and deeper circulation and between the horizontal and meridional overturning circulation are not well understood. Through this review, the remaining issues to be addressed in the future were identified. These issues included a connection between the changing properties of CW, DW, and BW, and their horizontal and overturning circulations; the linkage of deep and abyssal circulations to the upper circulation, including upper water transport from and into the Western Pacific Ocean; and processes underlying the temporal variability in the path and intensity of CW, DW, and BW.

• Annual Conference on Human and Language Technology
• /
• 2018.10a
• /
• pp.147-150
• /
• 2018

목적 지향 대화 시스템은 자연어 이해, 대화 관리자, 자연어 생성과 같은 세분화 모델들의 결합으로 이루어져있어 하위 모델에 대한 오류 전파에 취약하다. 이러한 문제점을 해결하기 위해 자연어 이해 모델과 대화 관리자를 하나의 네트워크로 구성하고 오류에 강건한 심층 Q 네트워크를 제안한다. 본 논문에서는 대화의 전체 흐름을 파악 할 수 있는 순환 신경망인 LSTM에 심층 Q 네트워크 적용한 심층 순환 Q 네트워크 기반 목적 지향 대화 시스템을 제안한다. 실험 결과, 제안한 심층 순환 Q 네트워크는 LSTM, 심층 Q 네트워크보다 각각 정밀도 1.0%p, 6.7%p 높은 성능을 보였다.

• The Journal of the Acoustical Society of Korea
• /
• v.38 no.3
• /
• pp.308-313
• /
• 2019

Parkinson's disease, one of the three major diseases in old age, has more than 70 % of patients with speech disorders, and recently, diagnostic methods of Parkinson's disease through speech signals have been devised. In this paper, we propose a method of diagnosis of Parkinson's disease based on deep residual gated recurrent neural network using speech features. In the proposed method, the speech features for diagnosing Parkinson's disease are selected and applied to the deep residual gated recurrent neural network to classify Parkinson's disease patients. The proposed deep residual gated recurrent neural network, an algorithm combining residual learning with deep gated recurrent neural network, has a higher recognition rate than the traditional method in Parkinson's disease diagnosis.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2011.05a
• /
• pp.78-78
• /
• 2011

수자원 확보를 위해 건설된 저수지는 내외부위의 여러 가지 환경영향으로 인해 수질문제가 종종 발생한다. 이런 수질문제에 대처하기 위해 다양한 형태의 수질개선 장치가 고안되어 적용되고 있으며, 물순환장치도 그 중에 하나이다. 물순환장치는 수층에 산소를 공급하고 표층의 조류제어, 이취미 제거 등 다양한 목적으로 사용된다. 본 연구에서는 저수지에 설치된 대류식 물순환장치의 수질개선 효과를 평가해 보았다. 수체가 큰 저수지에서의 조사는 매우 어려운 점이 있으나, 조사방법은 물순환장치로 부터 일정거리에서 수심별 수질조사를 실시하였으며, 물순환장치가 가동되는 지역과 장치가 없는 지역에 대한 조사를 실시하여 그 차이 만큼을 수질개선 효과로 평가하였다. 현장조사는 물순환장치로부터 1m, 3m, 5m, 7m, 10m, 13m, 15m 떨어진 지점에서 유속과 수질조사를 실시하였으며, 유속은 표면유속만 측정하였고 수질은 수표면에서부터 저수지 바닥까지 조사를 실시하였다. 조사기간 동안 수체의 물리적 조건은 계절에 따라 다양한 변화를 보였으나, 양수된 저층수의 흐름은 수평방향으로 흘러들어가면서 온도변화에 따라 혼합되거나 하강하는 것으로 나타났다. 심층 혐기성층이 발달된 상황에서 물순환장치의 가동은 물순환장치 흡입구 근처의 혐기성층을 흡입하여 표층으로 확산시켰고 이로인해 주변의 수체가 물순환장치 흡입구로 밀려오면서 심층의 혐기성층이 약화되는 현상을 나타냈다. 조류가 발생한 기간중에 물순환장치에 의한 조류제어 효과는 심층수가 상승하여 확산되는 수평방향으로 반경 약 10m 지점까지 조류농도가 낮아지는 것을 확인할 수 있었으며 조류농도가 낮아지는 기간에는 그 영향범위가 조금 적어지는 것으로 나타났다. 이러한 결과로 볼때 저수지에 설치되어 운영되는 대류식 물순환장치는 심층의 혐기화 개선과 표층의 조류제어에 효과가 있는 것으로 판단할수 있으나, 조류제어에 대해서는 좀 더 심도있는 조사가 필요할 것으로 판단되었다.

• Journal of the Korean earth science society
• /
• v.23 no.3
• /
• pp.259-269
• /
• 2002

Based on sedimentary structures, degree of bioturbation, and internal erosional layers, the deep-sea core sediments in the East Sea (Ulleung and Yamato basins) and the Northwestern Pacific Ocean (Shikoku Basin) can be divided into two parts (upper and lower) with the boundary of around 10,000 years B.P. in age. The upper part of core KT94-10 from Shikoku Basin is characterized by low sedimentation rate, internal erosion layer, high degree of bioturbation and cross-lamination structures. It can be interpreted as the bottom-current deposits which show some different characteristics from turbidite or hemipelagic sediment. However, its lower part consists of highly bioturbated, massive mud, suggesting that it be not related to the influence of bottom current. On the other hand, the cores in Ulleung and Yamato basins do not show any evidence of bottom-current deposits: their upper parts consist of bioturbated mud, and lower parts are characterized by laminated mud with pyrite filaments, indicating anaerobic condition. Consequently, these sedimentological characteristics suggest that deep-sea circulation would be changed from slow-moving to fast-moving one at this bounding time commonly in the Northwestern Pacific Ocean and the East Sea. Also, even in the same time, the deep-sea circulation in the Northwestern Pacific area would be relatively faster than that in the East Sea.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2017.04a
• /
• pp.892-895
• /
• 2017

본 논문에서는 비분할 비디오로부터 이 비디오에 담긴 사람의 행동을 효과적으로 탐지해내기 위한 심층 신경망 모델을 제안한다. 일반적으로 비디오에서 사람의 행동을 탐지해내는 작업은 크게 비디오에서 행동 탐지에 효과적인 특징들을 추출해내는 과정과 이 특징들을 토대로 비디오에 담긴 행동을 탐지해내는 과정을 포함한다. 본 논문에서는 특징 추출 과정과 행동 탐지 과정에 이용할 심층 신경망 모델을 제시한다. 특히 비디오로부터 각 행동별 시간적, 공간적 패턴을 잘 표현할 수 있는 특징들을 추출해내기 위해서는 C3D 및 I-ResNet 합성곱 신경망 모델을 이용하고, 시계열 특징 벡터들로부터 행동을 자동 판별해내기 위해서는 양방향 BI-LSTM 순환 신경망 모델을 이용한다. 대용량의 공개 벤치 마크 데이터 집합인 ActivityNet 비디오 데이터를 이용한 실험을 통해, 본 논문에서 제안하는 심층 신경망 모델의 성능과 효과를 확인할 수 있었다.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2016.10a
• /
• pp.617-620
• /
• 2016

본 논문에서는 이미지 캡션 생성과 모델 전이에 효과적인 심층 신경망 모델을 제시한다. 본 모델은 멀티 모달 순환 신경망 모델의 하나로서, 이미지로부터 시각 정보를 추출하는 컨볼루션 신경망 층, 각 단어를 저차원의 특징으로 변환하는 임베딩 층, 캡션 문장 구조를 학습하는 순환 신경망 층, 시각 정보와 언어 정보를 결합하는 멀티 모달 층 등 총 5 개의 계층들로 구성된다. 특히 본 모델에서는 시퀀스 패턴 학습과 모델 전이에 우수한 LSTM 유닛을 이용하여 순환 신경망 층을 구성하고, 컨볼루션 신경망 층의 출력을 임베딩 층뿐만 아니라 멀티 모달 층에도 연결함으로써, 캡션 문장 생성을 위한 매 단계마다 이미지의 시각 정보를 이용할 수 있는 연결 구조를 가진다. Flickr8k, Flickr30k, MSCOCO 등의 공개 데이터 집합들을 이용한 다양한 비교 실험을 통해, 캡션의 정확도와 모델 전이의 효과 면에서 본 논문에서 제시한 멀티 모달 순환 신경망 모델의 우수성을 입증하였다.

• The Science & Technology
• /
• no.8 s.459
• /
• pp.36-41
• /
• 2007

• Journal of the Korean Society for Marine Environment & Energy
• /
• v.7 no.1
• /
• pp.42-46
• /
• 2004

Deep Ocean Water (DOW) is formed within restricted area including polar sea (high latitude) by cooling of surface seawater and globally circulating in the state of isolation from surface seawater. Although it is not as obvious as estuaries mixing, brine ground water is mixture of recirculated seawater and ground water. Seawater having high osmotic pressure infiltrates into an aquifer which is connected to the sea. In order to clarify the characteristics of deep ocean water and brine ground water, we investigated their origins, chemical compositions, water qualities and resources stabilities. While concentrations of stable isotopes (/sup 18/O and ²H) in seawater is 0‰, those in brine ground water is on meteoric water line or shifted toward oxygen line. It means that origin of brine ground water is different than that of deep ocean water. The ions dissolved in seawater (Na, Ca, Mg, K) are present in constant proportions to each other and to the total salt content of seawater. However deviations in ion proportions have been observed in some brine ground water. Some causes of these exception to the rule of constant proportions are due to many chemical reactions between periphery soil and ground water. While DOW has a large quantity of functional trace metals and biological affinity relative to brine ground water, DOW has relatively small amount of harmful bacteria and artificial pollutants.

• KIPS Transactions on Software and Data Engineering
• /
• v.6 no.4
• /
• pp.203-210
• /
• 2017

In this paper, we propose an effective neural network model for image caption generation and model transfer. This model is a kind of multi-modal recurrent neural network models. It consists of five distinct layers: a convolution neural network layer for extracting visual information from images, an embedding layer for converting each word into a low dimensional feature, a recurrent neural network layer for learning caption sentence structure, and a multi-modal layer for combining visual and language information. In this model, the recurrent neural network layer is constructed by LSTM units, which are well known to be effective for learning and transferring sequence patterns. Moreover, this model has a unique structure in which the output of the convolution neural network layer is linked not only to the input of the initial state of the recurrent neural network layer but also to the input of the multimodal layer, in order to make use of visual information extracted from the image at each recurrent step for generating the corresponding textual caption. Through various comparative experiments using open data sets such as Flickr8k, Flickr30k, and MSCOCO, we demonstrated the proposed multimodal recurrent neural network model has high performance in terms of caption accuracy and model transfer effect.