Uncertainty-weighted semi-supervised learning with dynamic entropy masking and Bhattacharyya-regularized loss

Abstract Semi-supervised learning (SSL) leverages labeled and unlabeled data for modern classification tasks. However, existing SSL approaches often underutilize moderately uncertain samples and may propagate errors from highly uncertain pseudo-labels, leading to suboptimal performance, in noisy and...

Ausführliche Beschreibung

Bibliografische Detailangaben

Hauptverfasser:	Mohammed Talal Ghazal, Jafar Tanha, Nasrin Shahi, SeyedEhsan Roshan
Format:	Artigo
Sprache:	Inglês
Veröffentlicht:	Nature Portfolio 2025-11-01
Schriftenreihe:	Scientific Reports
Schlagworte:	Semi-supervised learning Uncertainty-weighted training Bhattacharyya-Regularized divergence Dynamic masking Pseudo-labels
Online-Zugang:	https://doi.org/10.1038/s41598-025-30069-3
Tags:	Tag hinzufügen Keine Tags, Fügen Sie das erste Tag hinzu!