Journal
PATTERN RECOGNITION
Volume 130, Issue -, Pages -Publisher
ELSEVIER SCI LTD
DOI: 10.1016/j.patcog.2022.108777
Keywords
Dynamic mutual training; Inter-model disagreement; Noisy pseudo label; Semi-supervised learning
Funding
- National Key Research and Development Program of China [2019YFC1521104]
- National Natural Science Foundation of China [72192821, 61972157]
- Shanghai Science and Technology Commission [21511101200]
- Art major project of National Social Science Fund [I8ZD22]
- High-Level Talent Program for Innovation and Entrepreneurship (ShuangChuang Doctor) of Jiangsu Province [JSSCBS20211220]
- Wu Wenjun Honorary Doctoral Scholarship, AI Institute, Shanghai Jiao Tong University
- Shanghai Municipal Science and Technology Major Project [2021SHZDZX0102]
Ask authors/readers for more resources
This paper points out that leveraging inter-model disagreement between different models is key to locating pseudo label errors. By proposing Dynamic Mutual Training (DMT) with a dynamically re-weighted loss function, state-of-the-art performance has been achieved in tasks like image classification and semantic segmentation.
Recent semi-supervised learning methods use pseudo supervision as core idea, especially self-training methods that generate pseudo labels. However, pseudo labels are unreliable. Self-training methods usually rely on single model prediction confidence to filter low-confidence pseudo labels, thus remaining high-confidence errors and wasting many low-confidence correct labels. In this paper, we point out it is difficult for a model to counter its own errors. Instead, leveraging inter-model disagreement between different models is a key to locate pseudo label errors. With this new viewpoint, we propose mutual training between two different models by a dynamically re-weighted loss function, called Dynamic Mutual Training (DMT). We quantify inter-model disagreement by comparing predictions from two different models to dynamically re-weight loss in training, where a larger disagreement indicates a possible error and corresponds to a lower loss value. Extensive experiments show that DMT achieves state-ofthe-art performance in both image classification and semantic segmentation. Our codes are released at https://github.com/voldemortX/DST-CBC .(c) 2022 Elsevier Ltd. All rights reserved.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available