Separated and Independent Contrastive Semi-Supervised Learning for Imbalanced Datasets

Conventional semi-supervised learning (SSL) encounters challenges in effectively addressing issues associated with long-tail datasets, primarily stemming from imbalances within a dataset. Existing contrastive SSL methods typically rely on pseudo-labels generated from unlabeled data, which can be ina...

Full description

Saved in:
Bibliographic Details
Main Authors: Dongyoung Kim, Won-Sook Lee, Young-Woong Ko, Jeong-Gun Lee
Format: Article
Language:English
Published: IEEE 2025-01-01
Series:IEEE Access
Subjects:
Online Access:https://ieeexplore.ieee.org/document/11039800/
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849682784313933824
author Dongyoung Kim
Won-Sook Lee
Young-Woong Ko
Jeong-Gun Lee
author_facet Dongyoung Kim
Won-Sook Lee
Young-Woong Ko
Jeong-Gun Lee
author_sort Dongyoung Kim
collection DOAJ
description Conventional semi-supervised learning (SSL) encounters challenges in effectively addressing issues associated with long-tail datasets, primarily stemming from imbalances within a dataset. Existing contrastive SSL methods typically rely on pseudo-labels generated from unlabeled data, which can be inaccurate and introduce confirmation bias toward majority classes. To address this issue, we propose Separated and Independent Contrastive Semi-Supervised Learning (SICSSL), which applies supervised contrastive learning separately and independently to labeled and unlabeled samples. We validate SICSSL on four public benchmark datasets: CIFAR-10-LT, CIFAR-100-LT, STL-10-LT, and ImageNet-127. The results demonstrate consistent performance improvements over existing contrastive learning approaches. Furthermore, SICSSL is modular and easily integrates with recent SSL frameworks, including Auxiliary Balanced Classifier (ABC) and Adaptive Consistency Regularizer (ACR), enabling improved robustness in imbalanced semi-supervised settings. Source code is available at <uri>https://github.com/dongyyyyy/SICSSL</uri>
format Article
id doaj-art-33eccef4738b4f6798a72187916439b7
institution DOAJ
issn 2169-3536
language English
publishDate 2025-01-01
publisher IEEE
record_format Article
series IEEE Access
spelling doaj-art-33eccef4738b4f6798a72187916439b72025-08-20T03:24:06ZengIEEEIEEE Access2169-35362025-01-011310571210572310.1109/ACCESS.2025.358073811039800Separated and Independent Contrastive Semi-Supervised Learning for Imbalanced DatasetsDongyoung Kim0https://orcid.org/0000-0003-1998-7784Won-Sook Lee1Young-Woong Ko2https://orcid.org/0000-0003-3920-9314Jeong-Gun Lee3https://orcid.org/0000-0001-6218-4560Department of Computer Engineering, Hallym University, Chuncheon, South KoreaSchool of Electrical Engineering and Computer Science, University of Ottawa, Ottawa, ON, CanadaDepartment of Computer Engineering, Hallym University, Chuncheon, South KoreaDepartment of Computer Engineering, Hallym University, Chuncheon, South KoreaConventional semi-supervised learning (SSL) encounters challenges in effectively addressing issues associated with long-tail datasets, primarily stemming from imbalances within a dataset. Existing contrastive SSL methods typically rely on pseudo-labels generated from unlabeled data, which can be inaccurate and introduce confirmation bias toward majority classes. To address this issue, we propose Separated and Independent Contrastive Semi-Supervised Learning (SICSSL), which applies supervised contrastive learning separately and independently to labeled and unlabeled samples. We validate SICSSL on four public benchmark datasets: CIFAR-10-LT, CIFAR-100-LT, STL-10-LT, and ImageNet-127. The results demonstrate consistent performance improvements over existing contrastive learning approaches. Furthermore, SICSSL is modular and easily integrates with recent SSL frameworks, including Auxiliary Balanced Classifier (ABC) and Adaptive Consistency Regularizer (ACR), enabling improved robustness in imbalanced semi-supervised settings. Source code is available at <uri>https://github.com/dongyyyyy/SICSSL</uri>https://ieeexplore.ieee.org/document/11039800/Semi-supervised learningimbalanced semi-supervised learninglong-tail problemcontrastive learning
spellingShingle Dongyoung Kim
Won-Sook Lee
Young-Woong Ko
Jeong-Gun Lee
Separated and Independent Contrastive Semi-Supervised Learning for Imbalanced Datasets
IEEE Access
Semi-supervised learning
imbalanced semi-supervised learning
long-tail problem
contrastive learning
title Separated and Independent Contrastive Semi-Supervised Learning for Imbalanced Datasets
title_full Separated and Independent Contrastive Semi-Supervised Learning for Imbalanced Datasets
title_fullStr Separated and Independent Contrastive Semi-Supervised Learning for Imbalanced Datasets
title_full_unstemmed Separated and Independent Contrastive Semi-Supervised Learning for Imbalanced Datasets
title_short Separated and Independent Contrastive Semi-Supervised Learning for Imbalanced Datasets
title_sort separated and independent contrastive semi supervised learning for imbalanced datasets
topic Semi-supervised learning
imbalanced semi-supervised learning
long-tail problem
contrastive learning
url https://ieeexplore.ieee.org/document/11039800/
work_keys_str_mv AT dongyoungkim separatedandindependentcontrastivesemisupervisedlearningforimbalanceddatasets
AT wonsooklee separatedandindependentcontrastivesemisupervisedlearningforimbalanceddatasets
AT youngwoongko separatedandindependentcontrastivesemisupervisedlearningforimbalanceddatasets
AT jeonggunlee separatedandindependentcontrastivesemisupervisedlearningforimbalanceddatasets