Generalized locality preserving Maxi-Min Margin Machine.

Area of Science:

• Machine Learning
• Pattern Recognition
• Computer Vision

Background:

• Large margin classifiers are crucial in machine learning.
• Existing methods like Maxi-Min Margin Machine (M⁴) consider local and global views.
• Locality Preserving Projections (LPP) focus on preserving data manifold structure.

Purpose of the Study:

• To propose a novel large margin classifier, the Generalized Locality Preserving Maxi-Min Margin Machine (GLPM).
• To enhance both local and global learning capabilities in pattern recognition.
• To develop a more robust classifier compared to existing models.

Main Methods:

• Constructing within-class matrices using labeled training points in a supervised manner.
• Preserving intra-class manifold and global projection directions.
• Theoretical analysis of GLPM in relation to M⁴ and Locality Fisher Discriminant Analysis (LFDA).

Main Results:

• GLPM is shown to be a generalized M⁴ machine.
• GLPM demonstrates improved robustness by not assuming specific data distributions.
• Experimental results show GLPM outperforms M⁴ in both local and global learning.

Conclusions:

• GLPM offers a generalized and more robust approach to large margin classification.
• The method effectively preserves both local manifold structures and global projection information.
• GLPM shows significant advantages in performance over M⁴ on benchmark datasets.