Self Supervised Learning for Image Classification With Few Labels
Journal of Clinical Practice and Health Sciences · 25.06.2026 · 6 (1) · 109-115
DOI: 10.71465/jcphs59 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Clinical Practice and Health Sciences, online erschienen am 25.06.2026. Verfasst von Wei Ming Tan und 2 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0).
Abstract (Originalsprache)
Labeled data are often limited in image classification because annotation is costly and slow. This study proposes a self-supervised learning method for image classification with few labels. The method uses multi-view data augmentation, SimCLR and MoCo pretraining, and pseudo-label fine-tuning to learn useful features from unlabeled images. Experiments were conducted on CIFAR-100 and STL-10 with only 10% labeled data. Compared with supervised training under the same label setting, the proposed method improved classification accuracy by about 9.7%. The accuracy gap from full-supervised training was less than 2.5%, and normalized mutual information increased by about 13%. These results show that contrastive pretraining and pseudo-label learning can reduce the effect of label shortage and improve feature clustering. The method provides a practical way to improve image classification when labeled samples are limited.
Zitieren
APA
Tan, W. M., Lim, R., & Kumar, A. (2026). Self Supervised Learning for Image Classification With Few Labels. Journal of Clinical Practice and Health Sciences, 6(1), 109-115. https://doi.org/10.71465/jcphs59
Vancouver
Tan WM, Lim R, Kumar A. Self Supervised Learning for Image Classification With Few Labels. Journal of Clinical Practice and Health Sciences. 2026;6(1):109-115. doi:10.71465/jcphs59
BibTeX
@article{Tan2026,
author = {Tan, Wei Ming and Lim, Rachel and Kumar, Aisha},
title = {Self Supervised Learning for Image Classification With Few Labels},
journal = {Journal of Clinical Practice and Health Sciences},
year = {2026},
volume = {6},
number = {1},
pages = {109--115},
doi = {10.71465/jcphs59}
}
EUVIMED