Library

S-learning: a joint supervised and self-supervised approach for robust learning with noisy labels

Noisy labels, resulting from manual labeling errors or unreliable data sources, can lead to overfitting and reduced generalization performance in neural networks when used for direct training. Self-supervised learning, which operates independently of labels, can mitigate the negative effects of noisy labels. Inspired by joint training that combines supervised and self-supervised learning, we propose an efficient method named S-learning for dealing with noisy labels. This method dynamically generates a clean sample set for supervised learning via three selection strategies: small-loss selection within class, high-confidence selection, and high-similarity selection. Meanwhile, other samples are utilized for self-supervised learning. Comprehensive experiments conducted on both synthetic and real-world noisy datasets show that S-learning surpasses numerous state-of-the-art methods.

Face and Expression RecognitionComputer ScienceEngineering
Source

Paper

Full text

PDF

S-learning: a joint supervised and self-supervised approach for robust learning with noisy labels

OpenAlex · Face and Expression Recognition · 2025

Abstract

Noisy labels, resulting from manual labeling errors or unreliable data sources, can lead to overfitting and reduced generalization performance in neural networks when used for direct training. Self-supervised learning, which operates independently of labels, can mitigate the negative effects of noisy labels. Inspired by joint training that combines supervised and self-supervised learning, we propose an efficient method named S-learning for dealing with noisy labels. This method dynamically generates a clean sample set for supervised learning via three selection strategies: small-loss selection within class, high-confidence selection, and high-similarity selection. Meanwhile, other samples are utilized for self-supervised learning. Comprehensive experiments conducted on both synthetic and real-world noisy datasets show that S-learning surpasses numerous state-of-the-art methods.

Source of record · OpenAlex

Similar papers

Co-learningCheng Tan, Jun Xia, Lirong Wu et al.2021 In Library
Co-learning: Learning from Noisy Labels with Self-supervisionCheng Tan, Jun Xia, Lirong Wu et al.2021 · ACM Multimedia · 171 citations In Library
Self-semi-supervised Learning to Learn from NoisyLabeled DataJiacheng Wang, Yue Ma, Shuang Gao2020 · arXiv (Cornell University) · 1 citation In Library
A Robust Learning Framework Using Self-supervised Learning for Learning with Noisy LabelsRyota Higashimoto, Soh Yoshida, Mitsuji Muneyasu2022 · 2022 IEEE 11th Global Conference on Consumer Electronics (GCCE) · 0 citations In Library
Handling Noisy Labels for Robustly Learning from Self-Training Data for Low-Resource Sequence LabelingDebjit Paul, Mittul Singh, Michael A. Hedderich et al.2019 · North American Chapter of the Association for Computational Linguistics · 18 citations In Library
Robust Learning of Noisy Labels Based on Semi-Supervised TechniquesXinru Zhou, Menghan Gong, Zheng Li et al.2025 In Library
Combat Noisy Labels by Joint TrainingHui Li, Zhaodong Niu, Lingxiang Peng et al.2022 · 2022 2nd International Conference on Consumer Electronics and Computer Engineering (ICCECE) In Library
A Novel Self-Supervised Re-labeling Approach for Training with Noisy LabelsDevraj Mandal, S. Bharadwaj, Soma Biswas2019 · IEEE Workshop/Winter Conference on Applications of Computer Vision · 29 citations In Library
Self-Paced Robust Learning for Leveraging Clean Labels in Noisy DataXuchao Zhang, Xian Wu, Fanglan Chen et al.2020 · Proceedings of the AAAI Conference on Artificial Intelligence · 22 citations In Library
Learning with Noisy labels via Self-supervised Adversarial Noisy MaskingYuanpeng Tu, Boshen Zhang, Yuxi Li et al.2023 In Library
© 2026 NYSGPT2525 LLC