본문 바로가기 주메뉴 바로가기
22대 대한민국 국회 국회정보길라잡이 국회의원검색
회원가입 로그인
HOME

정보검색

소장정보 검색

국회도서관 홈으로 정보검색 소장정보 검색

결과 내 검색

동의어 포함

고급검색

인명/단체명

저자정보 상세정보
인명/단체명을 입력하세요.

키워드

  • 대표어

  • 외국어

-

목차보기

Title Page 1

Abstract 4

Contents 5

Ⅰ. Introduction 7

Ⅱ. Related Work 9

2.1. Recent Medical Imaging Studies 9

2.2. Domain Adaptation 9

2.3. Coreset Selection Algorithm 11

Ⅲ. Methodology 13

3.1. Problem statement 13

3.2. Iterative Framework for Adaptive Domain Adaptation 15

3.3. Clustering Uncertainty-weighted Embeddings for Coreset Selection 17

3.4. Minimax Entropy for Semi-Supervised Domain Adaptation 20

Ⅳ. Experiment 22

4.1. Data Description 22

4.2. Experiment Settings 23

4.3. Results Analysis 24

Ⅴ. Conclusion 34

References 35

List of Figures 6

Figure 1. Channel-wise pixel intensity distribution by camera devices and target patients 13

Figure 2. Performance evaluation on training dataset composition 14

Figure 3. Overview of the proposed method 15

Figure 4. Facial image preprocessing and target signs 22

Figure 5. Class ratio in the source and target domains 23

Figure 6. Performance trends for all experimental settings at each step 27

Figure 7. Performance trends on the proposed method for coreset selection ratio at each step 28

Figure 8. Performance trends across smartphone devices for the proposed method at each step (RC, IC... 29

Figure 9. Performance trends across smartphone devices for the proposed method at each step (CE, RE... 30

Figure 10. Performance trends across smartphone devices for the proposed method at each step (SE... 31

Figure 11. Training data composition for eyeball expressed signs at each step 32

Figure 12. Training data composition for eyelid expressed signs at each step 33

초록보기

 In online classification problems, ensuring high-quality training data in real-time data streams is chal-lenging, especially when facing domain shift and label uncertainty. To address this challenge, a novel iterative learning framework is proposed to enhance the quality of training data by utilizing active do-main adaptation techniques. Within the proposed iterative learning framework, the adaptive domain adaptation approach selectively retains only the most informative and reliable data points, while leaving the remaining uncertain labeled data unlabeled. This study demonstrates that intentionally leaving un-certain data unlabeled and applying semi-supervised learning outperforms applying supervised learning with uncertain labels. The effectiveness of the proposed methodology was validated in the Thyroid- Associated Orbitopathy detection problem using heterogeneous image streams.

추천서가 (다양한 추천 자료를 만나보세요)

권호기사보기

권호기사 목록 테이블로 기사명, 저자명, 페이지, 원문, 기사목차 순으로 되어있습니다.
기사명 저자명 페이지 원문 기사목차

권호

기사명 원문보기 기사목차
닫기