http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
- 中文 을 입력하시려면 zhongwen을 입력하시고 space를누르시면됩니다.
- 北京 을 입력하시려면 beijing을 입력하시고 space를 누르시면 됩니다.
RISS 인기검색어
- 검색키워드
- 저자 : 백재순(Jaesoon Baik) (검색결과 1 건)
- 무료
- 기관 내 무료
- 유료
-
백재순(Jaesoon Baik),김지후(Jihoo Kim),최준원(Jun Won Choi) 한국자동차공학회 2021 한국자동차공학회 부문종합 학술대회 Vol.2021 No.6
In this paper, we introduce a novel active learning algorithm using consistency regularization. Consistency regularization is the most successful approach in semi-supervised learning (SSL). The consistency regularization measures the discrepancy between the predictions from the student and teacher models. Minimizing the consistency regularization leads the student model to produce a robust output when the noisy input and model perturbation are given. Our intuition behind the proposed acquisition function is that the sample that produces inconsistent prediction between the student and teacher model should be labeled preferentially. In order to leverage the benefit of student-teacher structure in SSL, we adapt the stochastic weight averaging(SWA) for our acquisition function. We used the structure of the SWA ensemble model as a teacher model to achieve superior performance over the recent AL algorithm on CIFAR-100 dataset.
연관 검색어 추천
이 검색어로 많이 본 자료
활용도 높은 자료
