DML-PL: deep metric learning based pseudo-labeling framework for class imbalanced semi-supervised learning
Traditional class imbalanced learning algorithms require training data to be labeled, whereas semi-supervised learning algorithms assume that the class distribution is balanced. However, class imbalance and insufficient labeled data problems often coexist in practical real-world applications. Curren...
Published in: | Information Sciences |
---|---|
Main Authors: | , , |
Other Authors: | |
Format: | Article in Journal/Newspaper |
Language: | English |
Published: |
2023
|
Subjects: | |
Online Access: | https://hdl.handle.net/10356/170840 https://doi.org/10.1016/j.ins.2023.01.074 |
id |
ftnanyangtu:oai:dr.ntu.edu.sg:10356/170840 |
---|---|
record_format |
openpolar |
spelling |
ftnanyangtu:oai:dr.ntu.edu.sg:10356/170840 2023-11-05T03:41:36+01:00 DML-PL: deep metric learning based pseudo-labeling framework for class imbalanced semi-supervised learning Yan, Mi Hui, Siu Cheung Li, Ning School of Computer Science and Engineering 2023 https://hdl.handle.net/10356/170840 https://doi.org/10.1016/j.ins.2023.01.074 en eng Information Sciences Yan, M., Hui, S. C. & Li, N. (2023). DML-PL: deep metric learning based pseudo-labeling framework for class imbalanced semi-supervised learning. Information Sciences, 626, 641-657. https://dx.doi.org/10.1016/j.ins.2023.01.074 0020-0255 https://hdl.handle.net/10356/170840 doi:10.1016/j.ins.2023.01.074 2-s2.0-85149786935 626 641 657 © 2023 Elsevier Inc. All rights reserved. Engineering::Computer science and engineering Class Imbalanced Classification Semi-Supervised Learning Journal Article 2023 ftnanyangtu https://doi.org/10.1016/j.ins.2023.01.074 2023-10-06T00:22:46Z Traditional class imbalanced learning algorithms require training data to be labeled, whereas semi-supervised learning algorithms assume that the class distribution is balanced. However, class imbalance and insufficient labeled data problems often coexist in practical real-world applications. Currently, most existing class-imbalanced semi-supervised learning methods tackle these two problems separately, resulting in the trained model biased towards majority classes that have more data samples. In this study, we propose a deep metric learning based pseudo-labeling (DML-PL) framework that tackles both problems simultaneously for class-imbalanced semi-supervised learning. The proposed DML-PL framework comprises three modules: Deep Metric Learning, Pseudo-Labeling and Network Fine-tuning. An iterative self-training strategy is used to train the model multiple times. For each time of training, Deep Metric Learning trains a deep metric network to learn compact feature representations of labeled and unlabeled data. Pseudo-Labeling then generates reliable pseudo-labels for unlabeled data through labeled data clustering with nearest neighbors selection. Finally, Network Fine-tuning fine-tunes the deep metric network to generate better pseudo-labels in the subsequent training. The training ends when all the unlabeled data are pseudo-labeled. The proposed framework achieved state-of-the-art performance on the long-tailed CIFAR-10, CIFAR-100, and ImageNet127 benchmark datasets compared with baseline models. This study is supported by National Natural Science Foundation of China (62273230) and China Scholarship Council (No.202006230225) Article in Journal/Newspaper DML DR-NTU (Digital Repository at Nanyang Technological University, Singapore) Information Sciences 626 641 657 |
institution |
Open Polar |
collection |
DR-NTU (Digital Repository at Nanyang Technological University, Singapore) |
op_collection_id |
ftnanyangtu |
language |
English |
topic |
Engineering::Computer science and engineering Class Imbalanced Classification Semi-Supervised Learning |
spellingShingle |
Engineering::Computer science and engineering Class Imbalanced Classification Semi-Supervised Learning Yan, Mi Hui, Siu Cheung Li, Ning DML-PL: deep metric learning based pseudo-labeling framework for class imbalanced semi-supervised learning |
topic_facet |
Engineering::Computer science and engineering Class Imbalanced Classification Semi-Supervised Learning |
description |
Traditional class imbalanced learning algorithms require training data to be labeled, whereas semi-supervised learning algorithms assume that the class distribution is balanced. However, class imbalance and insufficient labeled data problems often coexist in practical real-world applications. Currently, most existing class-imbalanced semi-supervised learning methods tackle these two problems separately, resulting in the trained model biased towards majority classes that have more data samples. In this study, we propose a deep metric learning based pseudo-labeling (DML-PL) framework that tackles both problems simultaneously for class-imbalanced semi-supervised learning. The proposed DML-PL framework comprises three modules: Deep Metric Learning, Pseudo-Labeling and Network Fine-tuning. An iterative self-training strategy is used to train the model multiple times. For each time of training, Deep Metric Learning trains a deep metric network to learn compact feature representations of labeled and unlabeled data. Pseudo-Labeling then generates reliable pseudo-labels for unlabeled data through labeled data clustering with nearest neighbors selection. Finally, Network Fine-tuning fine-tunes the deep metric network to generate better pseudo-labels in the subsequent training. The training ends when all the unlabeled data are pseudo-labeled. The proposed framework achieved state-of-the-art performance on the long-tailed CIFAR-10, CIFAR-100, and ImageNet127 benchmark datasets compared with baseline models. This study is supported by National Natural Science Foundation of China (62273230) and China Scholarship Council (No.202006230225) |
author2 |
School of Computer Science and Engineering |
format |
Article in Journal/Newspaper |
author |
Yan, Mi Hui, Siu Cheung Li, Ning |
author_facet |
Yan, Mi Hui, Siu Cheung Li, Ning |
author_sort |
Yan, Mi |
title |
DML-PL: deep metric learning based pseudo-labeling framework for class imbalanced semi-supervised learning |
title_short |
DML-PL: deep metric learning based pseudo-labeling framework for class imbalanced semi-supervised learning |
title_full |
DML-PL: deep metric learning based pseudo-labeling framework for class imbalanced semi-supervised learning |
title_fullStr |
DML-PL: deep metric learning based pseudo-labeling framework for class imbalanced semi-supervised learning |
title_full_unstemmed |
DML-PL: deep metric learning based pseudo-labeling framework for class imbalanced semi-supervised learning |
title_sort |
dml-pl: deep metric learning based pseudo-labeling framework for class imbalanced semi-supervised learning |
publishDate |
2023 |
url |
https://hdl.handle.net/10356/170840 https://doi.org/10.1016/j.ins.2023.01.074 |
genre |
DML |
genre_facet |
DML |
op_relation |
Information Sciences Yan, M., Hui, S. C. & Li, N. (2023). DML-PL: deep metric learning based pseudo-labeling framework for class imbalanced semi-supervised learning. Information Sciences, 626, 641-657. https://dx.doi.org/10.1016/j.ins.2023.01.074 0020-0255 https://hdl.handle.net/10356/170840 doi:10.1016/j.ins.2023.01.074 2-s2.0-85149786935 626 641 657 |
op_rights |
© 2023 Elsevier Inc. All rights reserved. |
op_doi |
https://doi.org/10.1016/j.ins.2023.01.074 |
container_title |
Information Sciences |
container_volume |
626 |
container_start_page |
641 |
op_container_end_page |
657 |
_version_ |
1781698040858411008 |