Hierarchical graph visualization using neural networks

Area of Science:

• Computer Science
• Artificial Intelligence
• Data Visualization

Background:

• Hierarchical graph visualization aims to create easily interpretable 2D drawings.
• Traditional methods often use sequential, local optimization heuristics.
• These heuristics may not achieve optimal global solutions for readability criteria.

Purpose of the Study:

• To present a novel algorithm for hierarchical graph visualization.
• To utilize a Hopfield network for global optimization of visualization parameters.
• To simultaneously minimize edge crossings and total path length.

Main Methods:

• Developed an algorithm based on Hopfield network principles.
• Applied the algorithm to the hierarchical graph visualization problem.
• Compared performance against traditional barycentric and priority layout heuristics.

Main Results:

• The Hopfield network algorithm achieved crossing minimization comparable to the barycentric heuristic.
• Simultaneously reduced total path length by up to 50% compared to the priority layout heuristic.
• Produced 2D drawings with improved interpretability.

Conclusions:

• Hopfield networks offer a viable global optimization approach for graph visualization.
• This method effectively balances minimizing crossings and path length.
• The algorithm demonstrates significant improvements over existing heuristic methods.