Robust Graph-Based Semisupervised Learning for Noisy Labeled Data via Maximum Correntropy Criterion.

Area of Science:

• Machine Learning
• Computer Vision
• Data Science

Background:

• Semisupervised learning (SSL) effectively addresses limited labeled data by leveraging abundant unlabeled data.
• Traditional SSL methods often struggle with robustness when faced with significant labeling noise.
• Developing noise-robust SSL techniques is crucial for reliable model performance in real-world applications.

Purpose of the Study:

• To propose a novel robust graph-based semisupervised learning method.
• To enhance model generalization and robustness against labeling noise.
• To improve the performance of SSL in scenarios with imperfect labels.

Main Methods:

• A graph-based semisupervised learning framework is enhanced by incorporating supervised information into the regularizer.
• The maximum correntropy criterion is integrated to effectively suppress the impact of labeling noise.
• The method aims to ensure predicted cluster labels closely align with true labels.

Main Results:

• The proposed method demonstrates improved robustness in the presence of noisy labels.
• Experimental results confirm the strong generalization capabilities of the developed SSL model.
• The approach effectively mitigates the negative effects of labeling noise on model performance.

Conclusions:

• The novel robust graph-based SSL method offers a significant advancement for handling noisy datasets.
• The integration of maximum correntropy criterion proves effective in noise suppression.
• The proposed technique provides a reliable solution for learning strong generalization models with limited and noisy labeled data.