CPSL: A semi-supervised framework with class prototype-based modeling for combating noisy labels

doi:10.1016/j.patcog.2025.112318

CPSL: A semi-supervised framework with class prototype-based modeling for combating noisy labels

2025-08-21

0

PRE

AI

DOI：10.1016/j.patcog.2025.112318

Original

Original request for help

Save

Abstract

En 中文

• We propose a semi-supervised learning framework with class prototype-based modeling. • We design a detector using class prototypes to divide labeled and unlabeled subsets. • Knowledge distillation is applied to improve FixMatch and MixMatch modules. • Experiments show our method outperforms others under various noisy scenarios.

Keywords:

semi-supervised learning

class prototypes

knowledge distillation

noisy scenarios

FixMatch

MixMatch

CPSL: A semi-supervised framework with class prototype-based modeling for combating noisy labels

Abstract

Journal

Organization