Deep semi-supervised learning via dynamic anchor graph embedding in latent space.

Area of Science:

• Machine Learning
• Artificial Intelligence
• Data Science

Background:

• Deep semi-supervised graph embedding learning shows promise but struggles with noisy, large graphs.
• Existing methods are often misled by inaccurate graph connections and lead to large model sizes.

Purpose of the Study:

• To propose a novel deep semi-supervised algorithm for simultaneous graph embedding and node classification.
• To address issues of noisy graphs and large model sizes in existing methods.

Main Methods:

• Utilizes dynamic graph learning in neural network hidden layer space.
• Constructs an anchor graph using hidden layer features for neighborhood context sampling.
• Employs a consistency-constrained network and an embedding network, optimized jointly.

Main Results:

• The proposed method improves performance on graph embedding and node classification tasks.
• Outperforms state-of-the-art approaches on popular image and text datasets.
• Demonstrates effectiveness in handling noisy graph connections and reducing model size.

Conclusions:

• The novel dynamic graph learning approach enhances semi-supervised graph embedding and node classification.
• Offers a more robust and efficient solution for large-scale graph data.
• Provides a promising direction for future research in graph-based machine learning.