• 한국정보과학회 언어공학연구회:학술대회논문집(한글 및 한국어 정보처리)
• /
• 한국정보과학회언어공학연구회 2017년도 제29회 한글 및 한국어 정보처리 학술대회
• /
• pp.35-40
• /
• 2017

기계 독해(Machine reading comprehension)는 주어진 문맥을 이해하고, 질문에 적합한 답을 문맥 내에서 찾는 문제이다. Simple Recurrent Unit (SRU)은 Gated Recurrent Unit (GRU)등과 같이 neural gate를 이용하여 Recurrent Neural Network (RNN)에서 발생하는 vanishing gradient problem을 해결하고, gate 입력에서 이전 hidden state를 제거하여 GRU보다 속도를 향상시킨 모델이며, Self-matching Network는 R-Net 모델에서 사용된 것으로, 자기 자신의 RNN sequence에 대하여 어텐션 가중치 (attention weight)를 계산하여 비슷한 의미 문맥 정보를 볼 수 있기 때문에 상호참조해결과 유사한 효과를 볼 수 있다. 본 논문에서는 한국어 기계 독해 데이터 셋을 구축하고, 여러 층의 SRU를 이용한 Encoder에 Self-matching layer를 추가한 $S^2$-Net 모델을 제안한다. 실험 결과, 본 논문에서 제안한 $S^2$-Net 모델이 한국어 기계 독해 데이터 셋에서 EM 65.84%, F1 78.98%의 성능을 보였다.

• 한국어정보학회:학술대회논문집
• /
• 한국어정보학회 2017년도 제29회 한글및한국어정보처리학술대회
• /
• pp.35-40
• /
• 2017

기계 독해(Machine reading comprehension)는 주어진 문맥을 이해하고, 질문에 적합한 답을 문맥 내에서 찾는 문제이다. Simple Recurrent Unit (SRU)은 Gated Recurrent Unit (GRU)등과 같이 neural gate를 이용하여 Recurrent Neural Network (RNN)에서 발생하는 vanishing gradient problem을 해결하고, gate 입력에서 이전 hidden state를 제거하여 GRU보다 속도를 향상시킨 모델이며, Self-matching Network는 R-Net 모델에서 사용된 것으로, 자기 자신의 RNN sequence에 대하여 어텐션 가중치 (attention weight)를 계산하여 비슷한 의미 문맥 정보를 볼 수 있기 때문에 상호참조해결과 유사한 효과를 볼 수 있다. 본 논문에서는 한국어 기계 독해 데이터 셋을 구축하고, 여러 층의 SRU를 이용한 Encoder에 Self-matching layer를 추가한 $S^2$-Net 모델을 제안한다. 실험 결과, 본 논문에서 제안한 $S^2$-Net 모델이 한국어 기계 독해 데이터 셋에서 EM 65.84%, F1 78.98%의 성능을 보였다.

• ETRI Journal
• /
• 제41권3호
• /
• pp.371-382
• /
• 2019

Machine reading comprehension is the task of understanding a given context and finding the correct response in that context. A simple recurrent unit (SRU) is a model that solves the vanishing gradient problem in a recurrent neural network (RNN) using a neural gate, such as a gated recurrent unit (GRU) and long short-term memory (LSTM); moreover, it removes the previous hidden state from the input gate to improve the speed compared to GRU and LSTM. A self-matching network, used in R-Net, can have a similar effect to coreference resolution because the self-matching network can obtain context information of a similar meaning by calculating the attention weight for its own RNN sequence. In this paper, we construct a dataset for Korean machine reading comprehension and propose an $S^2-Net$ model that adds a self-matching layer to an encoder RNN using multilayer SRU. The experimental results show that the proposed $S^2-Net$ model has performance of single 68.82% EM and 81.25% F1, and ensemble 70.81% EM, 82.48% F1 in the Korean machine reading comprehension test dataset, and has single 71.30% EM and 80.37% F1 and ensemble 73.29% EM and 81.54% F1 performance in the SQuAD dev dataset.

• Journal of Digestive Cancer Research
• /
• 제5권1호
• /
• pp.37-43
• /
• 2017

Laparoscopic cholecystectomy is of one the most common general surgical operations performed today. Concomitant choledocholithiasis occurs in roughly 10-20% of patients with symptomatic gallstones. Laparoscopic bile duct exploration (LBDE) offers a single-stage minimally-invasive solution to the management of choledocholithiasis. LBDE may be performed either via the transcystic route or via laparoscopic choledochotomy. A number of strategies to improve success are available to the surgeon to help in the problem of complicated choledocholithiasis, these range from simple maneuvers to the use of laser or mechanical lithotriptors. With the advances in laparoscopic surgery, it is also possible to handle complex surgical conditions such as Mirizzi syndrome or recurrent pyogenic cholangitis laparoscopically, even though these have yet to be accepted as standard of care. Following laparoscopic choledochotomy, options for closure include: primary closure, closure over a T-tube, and closure over an endobiliary stent. T-tube placement has been associated with increased operating time and hospital length of stay compared to primary closure, with no significant differences in morbidity. Based on the available literature, LBDE appears comparable to ERCP with regards to procedural efficacy and morbidity. LBDE remains relevant to the general surgeon and is best viewed as being complementary to endoscopic therapy in the management of choledocholithiasis.

• Asian Pacific Journal of Cancer Prevention
• /
• 제15권2호
• /
• pp.1035-1039
• /
• 2014

Aim: To evaluate the operative, oncologic and obstetric outcomes of the loop electrosurgical excision procedure (LEEP) in cases with cervical neoplasia. Materials and Methods: A retrospective cohort study was conducted on patients who were suspected of cervical neoplasia and therefore undergoing LEEP at Siriraj Hospital, Mahidol University, Thailand, during 1995-2000. Outcome measures included operative complications in 407 LEEP patients and long-term outcomes in the 248 patients with cervical intraepithelial neoplasia (CIN) who were treated with only LEEP. Results: There were 407 patients undergoing LEEP; their mean age was $39.7{\pm}10.5$ years. The histopathology of LEEP specimens revealed that 89 patients (21.9%) had lesions ${\leq}CIN$ I, 295 patients (72.5%) had CIN II or III, and 23 patients (5.6%) had invasive lesions. Operative complications were found in 15 patients and included bleeding (n=9), and infection (n=7). After diagnostic LEEP, 133 patients underwent hysterectomy as the definite treatment for cervical neoplasia. Of 248 CIN patients who had LEEP only, seven (2.8%) had suffered recurrence after a median of 16 (range 6-93) months; one had CIN I, one had CIN II, and five had CIN III. All of these recurrent patients achieved remission on surgical treatment with re-LEEP (n=6) or simple hysterectomy (n=1). A significant factor affecting recurrent disease was the LEEP margin involved with the lesion (p=0.05). Kaplan-Meier analysis showed 5-year and 10-year disease-free survival (DFS) estimates of 99.9%. Twelve patients became pregnant a total of 14 times, resulting in 12 term deliveries and two miscarriages - one of which was due to an incompetent cervix. Conclusions: LEEP for patients with cervical neoplasia delivers favorable surgical, oncologic and obstetric outcomes.

• 한국수자원학회:학술대회논문집
• /
• 한국수자원학회 2021년도 학술발표회
• /
• pp.131-131
• /
• 2021

Deep learning models, especially those based on long short-term memory (LSTM), have presented their superiority in addressing time series data issues recently. This study aims to comprehensively evaluate the performance of deep learning models that belong to the supervised learning category in streamflow prediction. Therefore, six deep learning models-standard LSTM, standard gated recurrent unit (GRU), stacked LSTM, bidirectional LSTM (BiLSTM), feed-forward neural network (FFNN), and convolutional neural network (CNN) models-were of interest in this study. The Red River system, one of the largest river basins in Vietnam, was adopted as a case study. In addition, deep learning models were designed to forecast flowrate for one- and two-day ahead at Son Tay hydrological station on the Red River using a series of observed flowrate data at seven hydrological stations on three major river branches of the Red River system-Thao River, Da River, and Lo River-as the input data for training, validation, and testing. The comparison results have indicated that the four LSTM-based models exhibit significantly better performance and maintain stability than the FFNN and CNN models. Moreover, LSTM-based models may reach impressive predictions even in the presence of upstream reservoirs and dams. In the case of the stacked LSTM and BiLSTM models, the complexity of these models is not accompanied by performance improvement because their respective performance is not higher than the two standard models (LSTM and GRU). As a result, we realized that in the context of hydrological forecasting problems, simple architectural models such as LSTM and GRU (with one hidden layer) are sufficient to produce highly reliable forecasts while minimizing computation time because of the sequential data nature.

• Communications for Statistical Applications and Methods
• /
• 제28권4호
• /
• pp.329-338
• /
• 2021

Deep learning methods have been developed, used in various fields, and they have shown outstanding performances in many cases. Many studies predicted a daily stock return, a classic example of time-series data, using deep learning methods. We also tried to apply deep learning methods to Korea's stock market data. We used Korea's stock market index (KOSPI) and several individual stocks to forecast daily returns and directions. We compared several deep learning models with other machine learning methods, including random forest and XGBoost. In regression, long short term memory (LSTM) and gated recurrent unit (GRU) models are better than other prediction models. For the classification applications, there is no clear winner. However, even the best deep learning models cannot predict significantly better than the simple base model. We believe that it is challenging to predict daily stock return data even if we use the latest deep learning methods.

• 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.

• 한국수자원학회:학술대회논문집
• /
• 한국수자원학회 2022년도 학술발표회
• /
• pp.149-149
• /
• 2022

Accurate quantitative evaluation of baseflow contribution to streamflow is imperative to address seasonal drought vulnerability, flood occurrence and groundwater management concerns for efficient and sustainable water resources management in watersheds. Several baseflow separation algorithms using recursive filters, graphical method and tracer or chemical balance have been developed but resulting baseflow outputs always show wide variations, thereby making it hard to determine best separation technique. Therefore, the current global shift towards implementation of artificial intelligence (AI) in water resources is employed to compare the performance of deep learning models with conventional hydrograph separation techniques to quantify baseflow contribution to streamflow of Piney River watershed, Tennessee from 2001-2021. Streamflow values are obtained from the USGS station 03602500 and modeled to generate values of Baseflow Index (BI) using Web-based Hydrograph Analysis (WHAT) model. Annual and seasonal baseflow outputs from the traditional separation techniques are compared with results of Long Short Term Memory (LSTM) and simple Gated Recurrent Unit (GRU) models. The GRU model gave optimal BFI values during the four seasons with average NSE = 0.98, KGE = 0.97, r = 0.89 and future baseflow volumes are predicted. AI offers easier and more accurate approach to groundwater management and surface runoff modeling to create effective water policy frameworks for disaster management.

• 지능정보연구
• /
• 제25권3호
• /
• pp.43-62
• /
• 2019

한때, 이상 탐지 분야는 특정 데이터로부터 도출한 기초 통계량을 기반으로 이상 유무를 판단하는 방법이 지배적이었다. 이와 같은 방법론이 가능했던 이유는 과거엔 데이터의 차원이 단순하여 고전적 통계 방법이 효과적으로 작용할 수 있었기 때문이다. 하지만 빅데이터 시대에 접어들며 데이터의 속성이 복잡하게 변화함에 따라 더는 기존의 방식으로 산업 전반에 발생하는 데이터를 정확하게 분석, 예측하기 어렵게 되었다. 따라서 기계 학습 방법을 접목한 SVM, Decision Tree와 같은 모형을 활용하게 되었다. 하지만 지도 학습 기반의 모형은 훈련 데이터의 이상과 정상의 클래스 수가 비슷할 때만 테스트 과정에서 정확한 예측을 할 수 있다는 특수성이 있고 산업에서 생성되는 데이터는 대부분 정답 클래스가 불균형하기에 지도 학습 모형을 적용할 경우, 항상 예측되는 결과의 타당성이 부족하다는 문제점이 있다. 이러한 단점을 극복하고자 현재는 클래스 분포에 영향을 받지 않는 비지도 학습 기반의 모델을 바탕으로 이상 탐지 모형을 구성하여 실제 산업에 적용하기 위해 시행착오를 거치고 있다. 본 연구는 이러한 추세에 발맞춰 적대적 생성 신경망을 활용하여 이상 탐지하는 방법을 제안하고자 한다. 시퀀스 데이터를 학습시키기 위해 적대적 생성 신경망의 구조를 LSTM으로 구성하고 생성자의 LSTM은 2개의 층으로 각각 32차원과 64차원의 은닉유닛으로 구성, 판별자의 LSTM은 64차원의 은닉유닛으로 구성된 1개의 층을 사용하였다. 기존 시퀀스 데이터의 이상 탐지 논문에서는 이상 점수를 도출하는 과정에서 판별자가 실제데이터일 확률의 엔트로피 값을 사용하지만 본 논문에서는 자질 매칭 기법을 활용한 함수로 변경하여 이상 점수를 도출하였다. 또한, 잠재 변수를 최적화하는 과정을 LSTM으로 구성하여 모델 성능을 향상시킬 수 있었다. 변형된 형태의 적대적 생성 모델은 오토인코더의 비해 모든 실험의 경우에서 정밀도가 우세하였고 정확도 측면에서는 대략 7% 정도 높음을 확인할 수 있었다.