.png){kind=icon}
.png){kind=icon}
Return
PrioMatch: Semi-supervised learning guided by prior knowledge
.png){kind=icon}
PRE
AI
DOI：10.1016/j.neucom.2025.131538.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Current semi-supervised learning algorithms mainly employ pseudo-labeling and consistency regularization to provide supervised signals for unlabeled data. However, most existing algorithms focus on mining supervisory signals intrinsic to datasets, while overlooking the abundant external knowledge. Fortunately, the advent of pre-trained large-scale models, which are imbued with prior knowledge, holds the promise of providing higher-quality external supervision for training. In this paper, we propose an innovative algorithm called PrioMatch. Guided by the prior knowledge provided by pre-trained visual models, we introduce an adaptive pseudo-label generation mechanism along with a corresponding pseudo-label reliability assessment strategy. Furthermore, we propose a dual dynamic class-balanced threshold mechanism, which dynamically adjusts the model’s focus during training to maintain a balanced recognition capability for different classes. We extensively evaluate our method on multiple benchmarks, including both balanced and imbalanced settings. The results demonstrate that PrioMatch consistently outperforms existing state-of-the-art methods, achieving a significant enhancement in both training efficiency and model accuracy. Code is available at https://github.com/JiaQuan1203/PrioMatch .
Journal
IF:
6.5
Papers:
2.5W
Citations:
6.5W