State transitions unlock temporal memory in swarm-based reservoir computing

This paper studies swarm-based reservoir computing with a focus on temporal memory rather than single-task forecasting. We show that introducing simple state transitions between dispersed and clustered behavior unlocks substantial temporal memory in the swarm and yields a clear linear scaling law between memory capacity and swarm size across the tested regime.

The work also provides a practical GPU-ready implementation recipe and helps clarify when multi-agent collectives can function as useful computational substrates.

Read the paper at PeerJ

Recommended citation: Lund, T., Adams, A., Aubert-Kato, N., & Ikegami, T. (2026). State transitions unlock temporal memory in swarm-based reservoir computing. PeerJ Computer Science, 12:e3763. https://doi.org/10.7717/peerj-cs.3763