Distribution-free Bayesian regularized learning framework for semi-supervised learning.

Area of Science:

• Machine Learning
• Artificial Intelligence
• Data Science

Background:

• Accurate data distribution knowledge is crucial but often impractical in machine learning.
• Semi-supervised learning requires effective methods for handling limited labeled data.

Purpose of the Study:

• To propose a novel distribution-free Bayesian regularized learning framework for semi-supervised learning.
• To develop a robust and efficient machine learning model applicable to real-world problems.

Main Methods:

• Introduced the Hessian regularized twin minimax probability extreme learning machine (HRTMPELM).
• Utilized high separation probability assumption to construct non-parallel hyperplanes.
• Incorporated Hessian regularization for geometric distribution information and Chebyshev inequalities for global optimization.

Main Results:

• Demonstrated reliability and effectiveness across multiple datasets compared to existing methods.
• Achieved simplified and efficient learning due to the absence of hyperparameters.
• Successfully applied to Ningxia wolfberry quality detection, showcasing agricultural applications.

Conclusions:

• The proposed HRTMPELM framework offers a robust, efficient, and generalizable solution for semi-supervised learning.
• The distribution-free approach and hyperparameter-free design significantly enhance practical applicability.
• The framework shows promise for advancing machine learning in agriculture and other fields.