Learned emergence in selfish collective motion.

Area of Science:

• Computational Biology
• Artificial Intelligence
• Collective Behavior

Background:

• Agent-based models (ABMs) are common for studying collective motion but can be computationally complex.
• The biological accuracy of imposed rules in ABMs is often questioned.
• Reservoir computing (RC) offers a framework for state-updating, but its application to swarm learning is unexplored.

Purpose of the Study:

• To investigate the application of reservoir computing to learn movement rules in a swarm of selfish individuals.
• To determine if selfish, risk-averse movement can be learned using an echo state network (ESN).
• To explore a novel approach for discovering realistic movement rules in autonomous systems.

Main Methods:

• Utilized an echo state network (a type of reservoir computing) for learning.
• Generated data from an agent-based model simulating selfish individuals optimizing their domains.
• Focused on input and output state representation for successful learning.

Main Results:

• Demonstrated that selfish, risk-minimizing movement in a swarm can be learned using an echo state network.
• Showcased the effectiveness of RC in capturing complex emergent behaviors from simple individual motivations.
• Validated the approach using data from a pre-existing agent-based model.

Conclusions:

• Reservoir computing provides a viable framework for learning complex collective movement rules.
• This suggests that sophisticated neural networks, like brains, could learn similar behaviors.
• Opens new avenues for realistic modeling of autonomous individual movement and swarm dynamics.