Maximum margin semi-supervised learning with irrelevant data

Area of Science:

• Machine Learning
• Artificial Intelligence
• Data Science

Background:

• Semi-supervised learning (SSL) typically assumes unlabeled data aligns with labeled data distributions.
• A significant challenge in SSL arises when unlabeled data contains irrelevant information, potentially degrading model performance.
• Existing SSL models often struggle to effectively mitigate the impact of such irrelevant data.

Purpose of the Study:

• To develop a robust semi-supervised classification model capable of handling irrelevant unlabeled data.
• To introduce the tri-class support vector machine (3C-SVM), a novel maximum margin model designed for this challenging scenario.
• To provide theoretical analysis and demonstrate the practical effectiveness of the proposed 3C-SVM.

Main Methods:

• The study proposes a tri-class support vector machine (3C-SVM) model that leverages both labeled and unlabeled data.
• 3C-SVM utilizes logistic and maximum entropy principles to reduce the influence of irrelevant unlabeled data.
• The model is solved efficiently using a sequence of quadratic programming subproblems, derived from a concave-convex programming approach.

Main Results:

• The proposed 3C-SVM effectively distinguishes between relevant and irrelevant unlabeled data, approaching an ideal classifier.
• Theoretical analysis confirms conditions under which irrelevant data can aid hyperplane optimization.
• 3C-SVM is shown to be a generalized framework, encompassing standard SVMs, S(3)VMs, and U-SVMs as special cases.

Conclusions:

• The 3C-SVM offers a powerful and flexible approach to semi-supervised classification in the presence of irrelevant unlabeled data.
• The model demonstrates superior effectiveness and efficiency compared to existing methods through extensive experiments.
• 3C-SVM provides a unified framework for maximum margin models, advancing the field of semi-supervised learning.