Multi-Agent Reinforcement Learning for Cooperative Large Language Model Collaboration
Global Media and Social Sciences Research Journal · 10.03.2026 · 7 (1) · 225-233
DOI: 10.71465/gmssrj167 ↗Kurz gefasst
Fachartikel in der Zeitschrift Global Media and Social Sciences Research Journal, online erschienen am 10.03.2026. Verfasst von Thomas J. Bennett und 2 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
This study investigates cooperative LLM collaboration through a multi-agent reinforcement learning (MARL) framework, where agents are jointly optimized using a shared group reward rather than independent objectives. A multi-agent GRPO (MAGRPO) algorithm is proposed to stabilize learning under decentralized execution. Evaluation is performed on 4 collaborative benchmarks covering reasoning, writing, and code synthesis, totaling 5,200 training episodes. Results show that MAGRPO improves average task reward by 27.6%, reduces inter-agent contradiction rates by 31.2%, and accelerates convergence by 1.8× compared to independent RL fine-tuning.
Zitieren
APA
Bennett, T. J., O’Neill, S. K., & Harding, L. M. (2026). Multi-Agent Reinforcement Learning for Cooperative Large Language Model Collaboration. Global Media and Social Sciences Research Journal, 7(1), 225-233. https://doi.org/10.71465/gmssrj167
Vancouver
Bennett TJ, O’Neill SK, Harding LM. Multi-Agent Reinforcement Learning for Cooperative Large Language Model Collaboration. Global Media and Social Sciences Research Journal. 2026;7(1):225-233. doi:10.71465/gmssrj167
BibTeX
@article{Bennett2026,
author = {Bennett, Thomas J. and O’Neill, Samuel K. and Harding, Laura M.},
title = {Multi-Agent Reinforcement Learning for Cooperative Large Language Model Collaboration},
journal = {Global Media and Social Sciences Research Journal},
year = {2026},
volume = {7},
number = {1},
pages = {225--233},
doi = {10.71465/gmssrj167}
}
EUVIMED