Fast Semisupervised Learning With Bipartite Graph for Large-Scale Data.

Area of Science:

• Computer Vision
• Machine Learning
• Data Science

Background:

• Labeled data is scarce, time-consuming, and expensive to acquire for real-world applications.
• Graph-based semi-supervised learning (GSSL) offers high accuracy but suffers from high computational complexity with large datasets.
• Existing GSSL methods often require powerful computing platforms, limiting their scalability.

Purpose of the Study:

• To propose a novel, efficient, and scalable semi-supervised learning method for computer vision.
• To address the computational complexity and scalability issues of traditional GSSL methods.
• To improve classification accuracy while reducing computational cost.

Main Methods:

• Constructing a parameter-insensitive, scale-invariant bipartite graph between original data and anchor points.
• Inferring labels for original data and anchors through graph-based label propagation.
• Extending the algorithm for out-of-sample data handling in large-scale scenarios using inferred anchor labels.

Main Results:

• The proposed bipartite GSSL normalized (BGSSL-normalized) method reduces computational complexity to O(ndm+nm^2).
• This represents a significant improvement over traditional GSSL methods with complexity O(n^2d+n^3).
• Experimental results on public datasets demonstrate superior classification accuracy and reduced time costs.

Conclusions:

• The BGSSL-normalized method effectively handles large-scale data in computer vision and machine learning.
• It offers a computationally efficient and accurate alternative to existing GSSL approaches.
• The method provides a scalable solution for semi-supervised learning tasks where data is limited.