Enhancing swarms' durability to threats via graph signal processing and graph-neural-network-based generative
Jonathan Karin1, Zoe Piran1, Mor Nitzan2
1Hebrew University of Jerusalem, School of Computer Science and Engineering, The , Jerusalem, Israel.
Physical Review. E
|January 21, 2026
Summary
This study models swarms as graphs to analyze external threats, revealing a detectability-durability trade-off crucial for swarm stability. A new generative model, SwaGen, optimizes this trade-off for designing robust artificial swarms.
Area of Science:
- Complex Systems
- Graph Theory
- Robotics
Background:
- Swarms in nature and engineering exhibit complex dynamics.
- Previous research overlooked external perturbations' impact on swarm stability.
- Understanding swarm resilience to threats is critical for both natural and artificial systems.
Purpose of the Study:
- To investigate the impact of external perturbations on swarm stability.
- To uncover the relationship between swarm detectability and durability against threats.
- To develop a generative model for designing resilient swarms.
Main Methods:
- Modeling swarms as graphs and applying graph signal processing.
- Analyzing predation as a perturbation signal on swarm graphs.
- Developing SwaGen, a graph neural network-based generative model for swarm optimization.
Main Results:
- Identified a detectability-durability trade-off influencing swarm resilience.
- Provided theoretical and empirical evidence linking this trade-off to spatial configuration.
- SwaGen successfully generated unique spatial configurations optimizing the trade-off.
Conclusions:
- External perturbations significantly affect swarm stability.
- The detectability-durability trade-off is a key factor in swarm resilience.
- SwaGen offers a novel approach for designing robust artificial swarms and understanding natural swarm dynamics.
