• 제목/요약/키워드: dropout

검색결과 264건 처리시간 0.027초

CNN기초로 세 가지 방법을 이용한 감정 표정 비교분석 (Comparative Analysis for Emotion Expression Using Three Methods Based by CNN)

  • 양창희;박규섭;김영섭;이용환
    • 반도체디스플레이기술학회지
    • /
    • 제19권4호
    • /
    • pp.65-70
    • /
    • 2020
  • CNN's technologies that represent emotional detection include primitive CNN algorithms, deployment normalization, and drop-off. We present the methods and data of the three experiments in this paper. The training database and the test database are set up differently. The first experiment is to extract emotions using Batch Normalization, which complemented the shortcomings of distribution. The second experiment is to extract emotions using Dropout, which is used for rapid computation. The third experiment uses CNN using convolution and maxpooling. All three results show a low detection rate, To supplement these problems, We will develop a deep learning algorithm using feature extraction method specialized in image processing field.

Improving Accuracy of Instance Segmentation of Teeth

  • Jongjin Park
    • International Journal of Internet, Broadcasting and Communication
    • /
    • 제16권1호
    • /
    • pp.280-286
    • /
    • 2024
  • In this paper, layered UNet with warmup and dropout tricks was used to segment teeth instantly by using data labeled for each individual tooth and increase performance of the result. The layered UNet proposed before showed very good performance in tooth segmentation without distinguishing tooth number. To do instance segmentation of teeth, we labeled teeth CBCT data according to tooth numbering system which is devised by FDI World Dental Federation notation. Colors for labeled teeth are like AI-Hub teeth dataset. Simulation results show that layered UNet does also segment very well for each tooth distinguishing tooth number by color. Layered UNet model using warmup trick was the best with IoU values of 0.80 and 0.77 for training, validation data. To increase the performance of instance segmentation of teeth, we need more labeled data later. The results of this paper can be used to develop medical software that requires tooth recognition, such as orthodontic treatment, wisdom tooth extraction, and implant surgery.

양방향 LSTM과 데이터 조합탐색 및 딥러닝 관련 기법을 활용한 철근 가격 단기예측에 관한 실험적 연구 (Experimental Study on the Short-Term Prediction of Rebar Price using Bidirectional LSTM with Data Combination and Deep Learning Related Techniques)

  • 이용성;김경환
    • 한국건설관리학회논문집
    • /
    • 제21권6호
    • /
    • pp.38-45
    • /
    • 2020
  • 본 연구는 양방향 LSTM, Random Search, 데이터 조합, Dropout을 이용한 철근 가격 단기예측 딥러닝 모델을 개발하는 체계적인 절차를 제시한다. 일반적으로 사용자가 직관적으로 이러한 값을 결정하여 예측성능이 우수한 결과를 탐색하는데 시간이 많이 걸리고 반복적인 시도를 하게 되는데, 이러한 시도로 찾아낸 결과가 우수하다고 보장할 수 없다. 본 연구에서 제시하는 제안된 접근방식으로 단기 가격예측의 평균 정확도는 약 98.32%이다. 그리고 이 방식은 철근 이외의 건축재료를 비롯하여 통계기반의 시계열 데이터로 가격을 예측하는 연구에서 본 연구에서 제시한 내용이 우수한 예측결과를 도출하기 위한 기초적 자료로 활용될 수 있을 것이다.

청소년상담데이터 기반 위기청소년 예측 (Youth Crisis Forecasting by Youth Counseling Data Analysis)

  • 이연희;천미경;송태민
    • 한국콘텐츠학회논문지
    • /
    • 제15권4호
    • /
    • pp.277-290
    • /
    • 2015
  • 본 연구는 비행 청소년이 학교폭력과 가정폭력, 학업중단, 자살이라는 위기로부터 일상의 건전한 청소년으로 회귀시키거나 선제적 개입 또는 사전 예방을 위해 실제로 위기에 내몰린 청소년을 분석하여 위기문제간의 연관된 패턴을 찾아내는데 목적이 있다. 연구 결과, 학업중단에 영향을 미치는 요인으로는 보호관찰, 범법, 흡연, 음주, 가출, 가정폭력_피해, 자살 등이 발견되었다. 특히 청소년이 가출을 해서 음주와 흡연을 하는 경우에는 학업중단의 위험이 그렇지 않은 청소년보다 2.76배 높은 것으로 예측되었고, 흡연보다 음주가 청소년의 학업중단에 더 크게 작용하는 것으로 밝혀졌다. 본 연구는 위기 청소년 문제를 해결하기 위해 당면한 문제에만 초점을 맞추기 보다는 잠재적 위기청소년의 가정과 학교, 지역사회 전반을 아우르는 복합적 위기관리가 필요함을 입증한 과학적 근거자료로서의 활용가치를 가진다.

Statistical Analysis on the trapping boundary of outer radiation belt during geosynchronous electron flux dropout : THEMIS observation

  • 황정아;이대영;김경찬;최은진;신대규;김진희;조정희
    • 천문학회보
    • /
    • 제37권1호
    • /
    • pp.90.2-90.2
    • /
    • 2012
  • Geosynchronous electron flux dropouts are most likely due to fast drift loss of the particles to the magnetopause (or equivalently, the "magnetopause shadowing effect"). A possible effect related to the drift loss is the radial diffusion of PSD due to gradient of PSD set by the drift loss effect at an outer L region. This possibly implies that the drift loss can affect the flux levels even inside the trapping boundary. We recently investigated the details of such diffusion process by solving the diffusion equation with a set of initial and boundary conditions set by the drift loss. Motivated by the simulation work, we have examined observationally the energy spectrum and pitch angle distribution near trapping boundary during the geosynchronous flux dropouts. For this work, we have first identified a list of geosynchronous flux dropout events for 2007-2010 from GOES satellite electron measurements and solar wind pressures observed by ACE satellite. We have then used the electron data from the Time History of Events and Macroscale Interactions during Substorms (THEMIS) spacecraft measurements to investigate the particle fluxes. The five THEMIS spacecraft sufficiently cover the inner magnetospheric regions near the equatorial plane and thus provide us with data of much higher spatial resolution. In this paper, we report the results of our investigations on the energy spectrum and pitch angle distribution near trapping boundary during the geosynchronous flux dropout events and discuss implications on the effects of the drift loss on the flux levels at inner L regions.

  • PDF

습식 부항 시술시 사혈량에 따른 부항 탈락 위험도 탐색 (Change in Risk of Dropout Due to Bleeding during Bloodletting-Cupping Therapy)

  • 김대혁;배은경;박정환;김소영;이상훈
    • Korean Journal of Acupuncture
    • /
    • 제35권1호
    • /
    • pp.41-45
    • /
    • 2018
  • Objectives : To investigate minimum pressure by verifying changes in pressure due to bleeding amount during bloodletting-cupping therapy. Methods : (1) We compared adhesion performance of four different cupping cups of same size: two disposable cupping cups(A, B) and two reusable cupping cups(A, B) each were vacuumed three times and kept in place for 10 minutes. (2) We vacuumed two different sized disposable cupping cups(A), size.1(InnerDiameter 48.8 mm) and size.3(InnerDiameter 39.1 mm), twice each(-200 mmHg) on silicon plate. We injected water and air at regular intervals in cupping cups by using a syringe, and then measured change of pressure in cupping cups and pressure at the time of dropout. Results : (1) Pressure reduction was $4.75{\pm}2.78%$ on average in the order of 'Disposable[A]>reusable[B]>Disposable[B]>reusable[A]', so that pressure retention performance of disposable cups can't be regarded as inferior to that of reusable cups. (2) Pressure of disposable cupping B(size.1) decreased by an average of -40.08 mmHg per 5 ml of water. At -24.8 mmHg, when 22 ml of water has been injected, cup has come off. Pressure of disposable cupping B(size. 3) decreased by an average of -99.4 mmHg per 5 ml of water. At -48.6 mmHg, when 13 ml of water was injected, cupping came off. Conclusions : Considering reduction rate of pressure due to water injection, in case of bleeding more than 15 ml, size.3 cup always comes off, therefore it needs to be re-operated at least once. Meanwhile, size.1 cup does not always come off in the same condition, depending on the initial pressure and therefore, re-operation may be considered.

$0.18{\mu}m$ CMOS 저 잡음 LDO 레귤레이터 (A Low-Noise Low Dropout Regulator in $0.18{\mu}m$ CMOS)

  • 한상원;김종식;원광호;신현철
    • 대한전자공학회논문지SD
    • /
    • 제46권6호
    • /
    • pp.52-57
    • /
    • 2009
  • 본 논문은 CMOS RFIC 단일 칩을 위한 Bandgap Voltage Reference와 이를 포함한 저 잡음 Low Dropout (LDO) Regulator 회로에 관한 것이다. 저 잡음을 위해 Bandgap Voltage Reference에 사용된 BJT 다이오드의 유효면적을 증가시켜야 함을 LDO의 잡음해석을 통해 나타내었다. 이를 위해 다이오드를 직렬 연결하여 실리콘의 실제면적은 최소화 하면서 다이오드의 유효면적을 증가시키는 방법을 적용하였고, 이를 통해 LDO의 출력잡음을 줄일 수 있음을 확인하였다. $0.18{\mu}m$ CMOS 공정으로 제작된 LDO는 입력전압이 2.2 V 에서 5 V 일때 1.8 V의 출력전압에서 최대 90 mA의 전류를 내보낼 수 있다. 측정 결과 Line regulation은 0.04%/V 이고 Load regulation은 0.45%를 얻었으며 출력 잡음 레벨은 100 Hz와 1 kHz offset에서 각각 479 nV/$^\surd{Hz}$와 186 nV/$^\surd{Hz}$의 우수한 성능을 얻었다.

Effects of and barriers to hospital-based pulmonary rehabilitation in patients with chronic obstructive pulmonary disease

  • Kim, Sang Hun;Jeong, Jong Hwa;Lee, Byeong Ju;Shin, Myung-Jun;Shin, Yong Beom
    • Physical Therapy Rehabilitation Science
    • /
    • 제9권2호
    • /
    • pp.82-89
    • /
    • 2020
  • Objective: The purpose of this study was to assess the effect of hospital-based pulmonary rehabilitation (PR) on exercise capacity and quality of life as well as barriers to participation in persons with chronic obstructive pulmonary disease (COPD) in South Korea. Design: One-group pretest-posttest design. Methods: A total of 14 patients were enrolled in this study in an 8-week PR program with two 60-minute sessions per week. The program included: flexibility exercises, breathing techniques, strengthening exercises, and aerobic exercises. The outcomes were defined as changes in the variables before and after the PR program. A change in the 6-minute walk distance (6MWD) was defined as the primary outcome, and changes in pulmonary function test, respiratory and grip strength, and the St. George's Respiratory Questionnaire (SGRQ) about quality-of-life results were secondary outcomes. A dropout was defined as missing >3 of the 16 sessions. Results: Patients who completed the program showed a significant improvement of 43.57±39.43 m in the 6MWD (p<0.05), but no significant differences were noted for the other function tests. The SGRQ showed a significant improvement in the activity and total score (p<0.05). The total dropout rate was 53.3%. Newly developed symptoms, exacerbation of COPD, transport problems, and lack of motivation were major barriers to PR. Conclusions: Our study showed that an 8-week hospital-based PR program improved exercise capacity and quality of life but had a high dropout rate in individuals with COPD. Since comprehensive PR has only recently been established in South Korea, patient motivation and education are critical.

셀프-캐스코드 구조를 적용한 LDO 레귤레이터 설계 (Design of Low Dropout Regulator using self-cascode structure)

  • 최성열;김영석
    • 한국정보통신학회논문지
    • /
    • 제22권7호
    • /
    • pp.993-1000
    • /
    • 2018
  • 본 논문에서는 셀프-캐스코드 구조를 이용한 LDO 레귤레이터를 제안하였다. 셀프-캐스코드 구조의 소스 측 MOSFET의 채널 길이를 조절하고, 드레인 측 MOSFET의 바디에 순방향 전압을 인가함으로써 최적화하였다. 오차 증폭기 입력 차동단의 셀프-캐스코드 구조는 높은 트랜스컨덕턴스를 가지도록, 출력단은 높은 출력 저항을 가지도록 최적화하였다. 제안 된 LDO 레귤레이터는 $0.18{\mu}m$ CMOS 공정을 사용하였고, SPECTERE를 이용하여 시뮬레이션 되었다. 제안 된 셀프-캐스코드 구조를 이용한 LDO 레귤레이터의 로드 레귤레이션은 0.03V/A로 기존 LDO의 0.29V/A보다 급격하게 개선되었다. 라인 레귤레이션은 2.23mV/V로 기존 회로보다 약 3배 향상되었다. 안정화 속도는 625ns로 기존 회로보다 346ns 개선되었다.

Application of Hot Start PCR Method in PCR-based Preimplantation Genetic Diagnosis

  • Kim, Sung-Ah;Kang, Moon-Joo;Kim, Hee-Sun;Oh, Sun-Kyung;Ku, Seung-Yup;Choi, Young-Min;Jun, Jong-Kwan;Moon, Shin-Yong
    • Journal of Genetic Medicine
    • /
    • 제9권1호
    • /
    • pp.11-16
    • /
    • 2012
  • Purpose: To determine a method to improve the efficacy and accuracy of preimplantation genetic diagnosis (PGD) - polymerase chain reaction (PCR), we compared hot start PCR and conventional multiplex nested PCR. Materials and Methods: This study was performed with single lymphocyte isolated from whole blood samples that were obtained from two couples with osteogenesis imperfecta (OI). We proceeded with conventional multiplex nested PCR and hot start PCR in which essential reaction components were physically removed, and we compared the amplification rate, allele dropout rate and nonspecific products. Afterward, we used selective method for PGD. Results: In the two couples, the respective amplification rate were 93.5% and 80.0% using conventional multiplex nested PCR and 95.5% and 92.0% using hot start PCR. The respective mean allele dropout rates for the two couples were 42.0% and 14.0% with conventional multiplex nested PCR and 36.0% and 6.0% with hot start PCR. Conclusion: The results demonstrate that the hot start PCR procedure provides higher amplification rates and lower allele dropout rate than the conventional method and that it decreased the nonspecific band in multiplex nested PCR. The hot start method is more efficient for analyzing a single blastomere in clinical PGD.