Uloženo v:
| Hlavní autor: | |
|---|---|
| Médium: | Recurso digital |
| Jazyk: | |
| Vydáno: |
Zenodo
2026
|
| Témata: | |
| On-line přístup: | https://doi.org/10.5281/zenodo.19325749 |
| Tagy: |
Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
|
| _version_ | 1866901355367497728 |
|---|---|
| author | Gomes, Renato Aparecido |
| author_facet | Gomes, Renato Aparecido |
| contents | Current multi-agent systems built on large language models rely on static orchestration patterns that fail to adapt when task demands shift dynamically. We present SYMBIONT (Symbiotic Multi-pattern Bio-intelligent Organism for Networked Tasks), a framework integrating eight biological swarm patterns into a unified organism for LLM agent coordination: mycorrhizal fungus networks (adaptive routing), slime mold topology optimization, ant caste polymorphism (5-caste agent specialization), honeybee waggle dance (collective decisions), termite mound architecture (stigmergy and homeostasis), starling murmuration (real-time reflexes), wolf/mole-rat governance (contextual leadership), and dolphin pod dynamics (ephemeral coalitions). The key design principle is functional inspiration, not structural replication. SYMBIONT is implemented in approximately 3,700 lines of Python with five agent castes mapping to LLM cost tiers, nine emergent laws, and a five-phase task execution lifecycle. Submitted to ANTS 2026 (15th International Conference on Swarm Intelligence, Darmstadt, Germany). |
| format | Recurso digital |
| id | zenodo_https___doi_org_10_5281_zenodo_19325749 |
| institution | Zenodo |
| language | |
| publishDate | 2026 |
| publisher | Zenodo |
| record_format | zenodo |
| spellingShingle | SYMBIONT: Unifying Eight Biological Swarm Patterns for LLM Agent Coordination Gomes, Renato Aparecido swarm intelligence multi-agent systems bio-inspired computing LLM agents collective intelligence self-organization SYMBIONT Current multi-agent systems built on large language models rely on static orchestration patterns that fail to adapt when task demands shift dynamically. We present SYMBIONT (Symbiotic Multi-pattern Bio-intelligent Organism for Networked Tasks), a framework integrating eight biological swarm patterns into a unified organism for LLM agent coordination: mycorrhizal fungus networks (adaptive routing), slime mold topology optimization, ant caste polymorphism (5-caste agent specialization), honeybee waggle dance (collective decisions), termite mound architecture (stigmergy and homeostasis), starling murmuration (real-time reflexes), wolf/mole-rat governance (contextual leadership), and dolphin pod dynamics (ephemeral coalitions). The key design principle is functional inspiration, not structural replication. SYMBIONT is implemented in approximately 3,700 lines of Python with five agent castes mapping to LLM cost tiers, nine emergent laws, and a five-phase task execution lifecycle. Submitted to ANTS 2026 (15th International Conference on Swarm Intelligence, Darmstadt, Germany). |
| title | SYMBIONT: Unifying Eight Biological Swarm Patterns for LLM Agent Coordination |
| topic | swarm intelligence multi-agent systems bio-inspired computing LLM agents collective intelligence self-organization SYMBIONT |
| url | https://doi.org/10.5281/zenodo.19325749 |