Related Experiment Videos
Marginal semi-supervised sub-manifold projections with informative constraints for dimensionality reduction and
Zhao Zhang1, Mingbo Zhao, Tommy W S Chow
1Department of Electronic Engineering, City University of Hong Kong, Tat Chee Avenue, Kowloon, Hong Kong. cszzhang@gmail.com
Summary
This study introduces two new semi-supervised dimensionality reduction algorithms, Marginal Semi-Supervised Sub-Manifold Projections (MS³MP) and orthogonal MS³MP (OMS³MP), for partial constrained data. These methods effectively select informative constraints and preserve data properties for improved performance.
Area of Science:
- Machine Learning
- Data Science
- Computer Vision
Background:
- Dimensionality reduction (DR) is crucial for analyzing high-dimensional data.
- Semi-supervised learning leverages limited labeled data alongside unlabeled data.
- Partial constrained data presents unique challenges for traditional DR methods.
Purpose of the Study:
- To propose novel semi-supervised DR algorithms for partial constrained data.
- To develop an effective technique for selecting informative pairwise constraints.
- To enhance the preservation of local data properties and discriminant structures.
Main Methods:
- Introduction of Marginal Semi-Supervised Sub-Manifold Projections (MS³MP) and orthogonal MS³MP (OMS³MP).
- Utilizing pairwise constraints (PC) and manifold scatters to guide DR.
- Developing an informative constraint selection technique for DR.
- Employing eigen-decomposition for analytic projection axes.
Main Results:
- Proposed MS³MP and OMS³MP algorithms effectively preserve local properties and discriminant structures.
- The novel constraint selection technique improves DR performance with consistent constraints.
- Sub-manifolds of different classes are successfully separated.
- Algorithms demonstrate promising results compared to state-of-the-art semi-supervised DR techniques.
Conclusions:
- The proposed MS³MP and OMS³MP algorithms offer a robust approach to semi-supervised DR with partial constrained data.
- The developed constraint selection method addresses a key challenge in PC-guided DR.
- The findings suggest significant improvements in data analysis and pattern recognition.