Semi-Supervised Regression via Hierarchical Confidence Screening and Neighbor-Guided Pseudo-Label Calibration
IEEE Access · 2026 · 14 · 52034-52053
DOI: 10.1109/access.2026.3680669 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Yunfei Mao und 1 weiteren Autor:innen. Die Arbeit nennt 55 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Semi-supervised regression enhances model performance by effectively utilizing unlabeled data during training. Most existing methods select only a small subset of high-confidence unlabeled samples based on single confidence metrics to participate in model training. However, these approaches suffer from incomplete reliability assessment and inefficient utilization of the majority of unlabeled data. In this paper, we propose a semi-supervised regression via hierarchical confidence screening and neighbor-guided pseudo-label calibration (S2RSC) algorithm that operates through two synergistic stages. First, we establish a consensus-driven selection mechanism where multiple diverse regressors evaluate prediction consistency, coupled with a safe labeling technique ensuring pseudo-label reliability. In this way, high-confidence unlabeled samples are reliably identified and incorporated into the labeled dataset for subsequent training. Second, we introduce an uncertainty-weighted unsupervised loss where Monte Carlo Dropout quantifies uncertainty to dynamically adjust sample contributions during training. Consequently, all remaining unlabeled data can be effectively utilized while mitigating the adverse effects of noisy pseudo-labels. Third, we propose a neighbor-guided pseudo-label calibration technique that leverages local labeled neighborhoods to adaptively calibrate pseudo-labels as a data augmentation strategy. This approach addresses the scarcity of augmentation methods for tabular data and enhances model robustness through effective consistency regularization. Experiments were performed on fifteen benchmark datasets from biomedical, business, ecology, physical, and life domains. The results demonstrate that S2RSC compares favorably against eight state-of-the-art semi-supervised regression algorithms across the majority of benchmark scenarios.
Zitieren
APA
Mao, Y., & Yao, B. (2026). Semi-Supervised Regression via Hierarchical Confidence Screening and Neighbor-Guided Pseudo-Label Calibration. IEEE Access, 14, 52034-52053. https://doi.org/10.1109/access.2026.3680669
Vancouver
Mao Y, Yao B. Semi-Supervised Regression via Hierarchical Confidence Screening and Neighbor-Guided Pseudo-Label Calibration. IEEE Access. 2026;14:52034-52053. doi:10.1109/access.2026.3680669
BibTeX
@article{Mao2026,
author = {Mao, Yunfei and Yao, Bowen},
title = {Semi-Supervised Regression via Hierarchical Confidence Screening and Neighbor-Guided Pseudo-Label Calibration},
journal = {IEEE Access},
year = {2026},
volume = {14},
pages = {52034--52053},
doi = {10.1109/access.2026.3680669}
}
EUVIMED