• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.18 no.3
• /
• pp.570-590
• /
• 2024

Breast cancer ranks among the most prevalent forms of malignancy and foremost cause of death by cancer worldwide. It is not preventable. Early and precise detection is the only remedy for lowering the rate of mortality and improving the probability of survival for victims. In contrast to present procedures, thermography aids in the early diagnosis of cancer and thereby saves lives. But the accuracy experiences detrimental impact by low sensitivity for small and deep tumours and the subjectivity by physicians in interpreting the images. Employing deep learning approaches for cancer detection can enhance the efficacy. This study explored the utilization of thermography in early identification of breast cancer with the use of a publicly released dataset known as the DMR-IR dataset. For this purpose, we employed a novel approach that entails the utilization of a pre-trained MobileNetV2 model and fine tuning it through transfer learning techniques. We created three models using MobileNetV2: one was a baseline transfer learning model with weights trained from ImageNet dataset, the second was a fine-tuned model with an adaptive learning rate, and the third utilized early stopping with callbacks during fine-tuning. The results showed that the proposed methods achieved average accuracy rates of 85.15%, 95.19%, and 98.69%, respectively, with various performance indicators such as precision, sensitivity and specificity also being investigated.

• Journal of Korean Academy of Fundamentals of Nursing
• /
• v.23 no.1
• /
• pp.61-72
• /
• 2016

Purpose: Purpose of this retrospective study was to investigate cognitive function, physical function, and problematic behaviors of elders who attended dementia daycare centers, and to identify reasons why they stopped using the center. Methods: Participants were 176 elders, 60 years or over, attending one of four dementia daycare centers in Incheon. Data were collected from center documents. Results: Mean age was 80.5 years. When admitted to the centers mean scores for the mini-mental status examination, activity of daily living, and instrumental activity of daily living (IADL) were 12.31, 9.53, and 25.09 respectively. Participants received day care service for an average of 17.98 months. The reasons for leaving the center were worsening dementia and health (40.2%), and problematic behaviors (20.1%). Conclusion: Results show that elders began to use day care services when their cognitive function and IADL had declined considerably. As the ultimate goal of dementia daycare service is to delay the worsening of cognitive capability and decreases in activities daily living, the effect of the service can be maximized when the service is provided as early as possible in the course of progressively severe dementia. Active promotion should be exerted in the community to encourage early use of this service.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.30 no.6C
• /
• pp.442-449
• /
• 2005

In this paper, we introduce three kinds of simplified high-speed decoding algorithms for turbo product decoder. First, A parallel decoder structure, the row and column decoders operate in parallel, is proposed. Second, HAD(Hard Decision Aided) algorithm is used for early-stopping algorithm. Lastly, P-Parallel TPC decoder is a parallel decoding scheme, processing P rows and P columns in parallel instead of decoding one by one as that in the original scheme.

• Annals of Clinical Neurophysiology
• /
• v.15 no.2
• /
• pp.37-41
• /
• 2013

Saccades are rapid eye movements that shift the line of sight between successive points of fixation. The cerebellum calibrates saccadic amplitude (dorsal vermis and fastigial nucleus) and the saccadic pulse-step match (flocculus) for optimal visuo-ocular motor behavior. Based on electrophysiology and the pharmacological inactivation studies, early activity in one fastigial nucleus could be important for accelerating the eyes at the beginning of a saccade, and the later activity in the other fastigial nucleus could be critical for stopping the eye on target, which is controlled by inhibitory projection from the dorsal vermis. The cerebellum could monitor a corollary discharge of the saccadic command and terminate the eye movement when it is calculated to be on target. The fastigial nucleus and dorsal vermis also participate in the adaptive control of saccadic accuracy.

• Proceedings of the Korean Society of Computer Information Conference
• /
• 2023.01a
• /
• pp.91-92
• /
• 2023

본 논문에서는 Yolov5와 Deepsort를 사용한 Tracking by detection을 구현하여 특정 영역을 통과하는 차량의 수를 집계하고, 각 차량의 추정속도를 계산하는 시스템을 구현한다. 실시간 객체 탐지 기능을 수행하는 Yolov5 모델의 학습에는 Kaggle의 개방 데이터인 '도요타 자동차 이미지'를 사용한다. 이미지 크기 640*640, 배치사이즈 16, Early stopping 플래그를 사용하여 학습했을때, Yolov5의 객체 탐지 성능은 정확도 98%, 정밀도 0.961, mAP 0.72을 보여주었다.

• Annual Conference of KIPS
• /
• 2024.05a
• /
• pp.780-783
• /
• 2024

본 논문에서는 지속적으로 최적화된 인공지능 모델을 적용하기 위한 방안으로 연합 학습(Federated Learning)을 활용한 접근법을 제시한다. 최근 다양한 산업 분야에서 인공지능 활용에 대한 필요성이 증가하고 있다. 금융과 같은 일부 산업은 강력한 보안, 높은 정확도, 규제 준수, 실시간 대응이 요구됨과 동시에 정적 시스템 환경 특성으로 적용된 인공지능 모델의 최적화가 어렵다. 이러한 환경적 한계 해결을 위하여, 연합 학습을 통한 모델의 최적화 방안을 제안한다. 연합 학습은 데이터 프라이버시를 유지하면서 모델의 지속적 최적화를 제공이 가능한 강력한 아키텍처이다. 그러나 연합 학습은 클라이언트와 중앙 서버의 반복적인 통신과 학습으로, 불필요한 자원에 대한 소요가 요구된다. 이러한 연합 학습의 단점 극복을 위하여, 주요도 높은 클라이언트의 선정 및 클라이언트와 중앙 서버의 조기 중단(early stopping) 전략을 통한 지속적, 효율적 최적화가 가능한 연합 학습 모델의 운영 전략을 제시한다.

• Tuberculosis and Respiratory Diseases
• /
• v.40 no.4
• /
• pp.404-409
• /
• 1993

Background: There are many retreatment failure patients admitted in National Kongju Tuberculosis Hospital. But there is not satisfactory treatment method for them at present. We think that more attentions and active measures for them are needed. Method: We reviewed sex and age, duration of illness, previously used antituberculosis drugs, drug resistance, extent of disease, reasons for early stopping or irregular medication and schooling of 50 retreatment failure patients admitted in National Kongju Tuberculosis Hospital from April 1992 to February 1993. Results: 1) The male to female ratio was 3:2 and 62% of the patients were between 21 and 40 years of age. 2) Twenty eight cases (56%) had the duration of illness over 10 years. 3) All cases had used most of the antituberculosis drugs. 4) Drug sensitivity test showed resistance to RMP in 46 cases (96%), INH in 40 cases (83%) and other drugs in 3-32 cases (6-67%). 5) Forty eight cases (96%) had far advanced disease on chest P-A film. 6) Twenty eight cases (56%) in primary chemotherapy and twenty one cases (42%) in retreatment had the histories of premature stopping or irregular ingestion of the drug. The reasons for premature stopping or irregular ingestion of the drug were as follows; in primary chemotherapy, 29 cases (75%) were due to 'having no symptoms', while in retreatment, 6 cases (29%) were due to 'having no symtoms', 6 cases (29%) were 'too, busy' and 3 cases (14%) were for 'financial problem'. 7) Twenty seven cases (54%) had at least graduated from high school. Conclusion: Greater efforts are needed to prevent tratment failure. More supports and admission treatment for retreatment failure patients are needed to prevent infection and to treat properly.

• Journal of Digital Convergence
• /
• v.14 no.8
• /
• pp.245-252
• /
• 2016

The purpose of this study is to make up for missing of weather informations from ASOS and AWS using artificial neural networks. We collected temperature, relative humidity and wind velocity for August during 5-yr (2011-2015) and sample designed artificial neural networks, assuming the Seoul weather station was missing. The result of sensitivity study on number of epoch shows that early stopping appeared at 2,000 epochs. Correlation between observation and prediction was higher than 0.6, especially temperature and humidity was higher than 0.9, 0.8 respectively. RMSE decreased gradually and training time increased exponentially with respect to increase of number of epochs. The predictability at 40 epoch was more than 80% effect on of improved results by the time the early stopping. It is expected to make it possible to use more detailed weather information via the rapid missing complemented by quick learning time within 2 seconds.

• Explosives and Blasting
• /
• v.38 no.4
• /
• pp.16-25
• /
• 2020

Multi-setting smart-investigation of the ground and large uncharged hole boring (MSP) method to reduce the blast-induced vibration in a tunnel excavation is carried out over 50m of long-distance boring in a horizontal direction and thus has been accompanied by deviations in boring alignment because of the heavy and one-directional rotation of the rod. Therefore, the deviation has been adjusted through the boring machine's variable setting rely on the previous construction records and expert's experience. However, the geological characteristics, machine conditions, and inexperienced workers have caused significant deviation from the target alignment. The excessive deviation from the boring target may cause a delay in the construction schedule and economic losses. A deep learning-based prediction model has been developed to discover an ideal initial setting of the MSP machine. Dropout, early stopping, pre-training techniques have been employed to prevent overfitting in the training phase and, significantly improved the prediction results. These results showed the high possibility of developing the model to suggest the boring machine's optimum initial setting. We expect that optimized setting guidelines can be further developed through the continuous addition of the data and the additional consideration of the other factors.

• Journal of Korean Tunnelling and Underground Space Association
• /
• v.11 no.2
• /
• pp.151-162
• /
• 2009

This study proposes a new methodology for quantitative rock classification in unsampled rock zone, which occupies the most of tunnel design area. This methodology is to train an ANN (artificial neural network) by using results from a drilling investigation combined with electric resistivity survey in sampled zone, and then apply the trained ANN to making a prediction of grade of rock classification in unsampled zone. The prediction is made at the center point of a shifting window by using a number of electric resistivity values within the window as input reference information. The ANN training in this study was carried out by the RPROP (Resilient backpropagation) training algorithm and Early-Stopping method for achieving a generalized training. The proposed methodology is then applied to generate a rock grade distribution on a real tunnel site where drilling investigation and resistivity survey were undertaken. The result from the ANN based prediction is compared with one from a conventional kriging method. In the comparison, the proposed ANN method shows a better agreement with the electric resistivity distribution obtained by field survey. And it is also seen that the proposed method produces a more realistic and more understandable rock grade distribution.