본문 바로가기

쓰기

 

Meta Pseudo Labels

by김누리 2021.06.23 18:02
발표자 김누리 
발표일자 2021-06-23 
저자 Hieu Pham, Zihang Dai, Qizhe Xie, Quoc V. Le 
학회명 Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2021. 
논문지  
We present Meta Pseudo Labels, a semi-supervised learning method that achieves a new state-of-the-art top-1 accuracy of 90.2% on ImageNet, which is 1.6% better than the existing state-of-the-art. Like Pseudo Labels, Meta Pseudo Labels has a teacher network to generate pseudo labels on unlabeled data to teach a student network. However, unlike Pseudo Labels where the teacher is kept fixed, in Meta Pseudo Labels, the teacher is constantly adapted by the feedback of how well the student performs on the labeled dataset. As a result, the teacher generates better pseudo labels to teach the student.

figure 1.png

    2021

      about confidence calibration
      2021.07.29
      발표자: 강석규     발표일자: 2021-07-29     저자: Chuan Guo, Geoff Pleiss, Yu Sun, Kilian Q. Weinberger     학회명: ICLR2017    
      NEVER GIVE UP: LEARNING DIRECTED EXPLORATION STRATEGIES
      2021.07.14
      발표자: 채경훈     발표일자: 2021-07-14     저자: Adrià Puigdomènech Badia∗ Pablo Sprechmann∗ Alex Vitvitskyi Daniel Guo Bilal Piot Steven Kapturowski Olivier Tieleman Martín Arjovsky Alexander Pritzel Andew Bolt Charles Blundell     학회명: ICLR