A Deep Generative Model for Reordering Adjacency Matrices.

Area of Science:

• Computer Science
• Graph Theory
• Machine Learning

Background:

• Graph visualization relies on adjacency matrices, where node ordering significantly impacts the display of graph characteristics.
• Current methods for finding optimal node orderings involve extensive trial-and-error, which is inefficient and challenging for users, especially novices.

Purpose of the Study:

• To develop a technique for effortless discovery of matrix reorderings for graph visualization.
• To address the limitations of manual, trial-and-error approaches in selecting effective node orderings for adjacency matrices.

Main Methods:

• Designed a generative model to learn a latent space encompassing diverse matrix reorderings of a given graph.
• Developed an intuitive user interface that maps the learned latent space, facilitating exploration of various reorderings.

Main Results:

• The generative model successfully learns a latent space that captures a wide variety of matrix reorderings.
• Quantitative and qualitative evaluations confirm the model's capability in generating diverse and useful reorderings.

Conclusions:

• This work presents a novel machine learning-driven approach to graph matrix visualization, moving beyond traditional algorithmic solutions.
• The proposed method empowers users to intuitively find desired matrix reorderings, enhancing the graph analysis process.