Improved Graph-based semi-supervised learning Schemes

TL;DR

The paper tackles semi-supervised classification with very limited labeled data by casting label propagation as PDEs on graphs. It introduces two main contributions: Modified Gaussian Random Fields (MGRF/IGRF) and Improved Poisson Learning (IPL), which bias propagation toward the graph's stationary distribution and apply a rank-one correction to stabilize updates. Across experiments on imbalanced and balanced datasets (e.g., Two-Moon, CIFAR-10, KEEL), the proposed methods yield higher accuracy and robustness than traditional graph-based SSL approaches, particularly when labels are scarce. The approach offers scalable improvements for real-world imbalanced data scenarios, combining efficiency with improved convergence and minority-class performance.

Abstract

In this work, we improve the accuracy of several known algorithms to address the classification of large datasets when few labels are available. Our framework lies in the realm of graph-based semi-supervised learning. With novel modifications on Gaussian Random Fields Learning and Poisson Learning algorithms, we increase the accuracy and create more robust algorithms. Experimental results demonstrate the efficiency and superiority of the proposed methods over conventional graph-based semi-supervised techniques, especially in the context of imbalanced datasets.

Figures (4)

• Figure 1: The classification on Two-Moon.
• Figure 2: The classification on Imbalanced Two-Moons.
• Figure 3: The classification on (two-circles ).
• Figure 4: The classification of Cifar-10.

Theorems & Definitions (5)

• Example 4.1
• Example 4.2
• Example 4.3
• Example 4.4
• Example 4.5