NeuroMAS: Multi-Agent Systems as Neural Networks with Joint Reinforcement Learning

Abstract

Multi-agent language systems are often built as hand-designed workflows, where agents are assigned semantic roles and communication protocols are specified in advance. We propose NeuroMAS, a method that first treats a multi-agent language system as a trainable and scalable neural-network-like architecture with LLM agents as nodes and intermediate textual signals as edges. In NeuroMAS, agent nodes are role-free but structure-aware: the topology only determines how information can flow in general, while reinforcement learning training determines how nodes communicate, specialize, and coordinate. This formulation shifts multi-agent design from workflow engineering toward architecture design, where depth, width, connectivity, and growth protocol become scalable sources of capability. Further, we provide a theoretical perspective showing why such modular textual computation is more parameter-efficient when tasks admit hierarchical decompositions. Experiments show that NeuroMAS improves significantly over both inference-time and trained multi-agent baselines. We further find that organizational scaling is path-dependent: larger systems can be challenging to train from scratch, but become feasible when grown progressively from smaller trained systems. These results suggest that learned neural multi-agent systems are a promising scaling axis for LLMs.

Bayesian Neural Networks via Dirichlet Process Priors on the Data-Generating Distribution for Randomized Value Functions in Deep Reinforcement Learning
Sumit Vashishtha
arXiv20 Oct 2026PreprintCC BY 4.0Full textStatisticsarXiv
Data-Driven On/Off Topology Optimization of Electromagnetic Devices Based on Graph Physics-Informed Neural Networks and Physics-Guided Reinforcement Learning
Xiaoyu Liu, Yihan Si, Wei Xiang +1
Mathematics9 Oct 2026Peer-reviewedCC BY 4.0General mathematicsCrossref
Efficient Joint Trajectory Optimization and Resource Allocation for Secure UAV Networks: A Deep Reinforcement Learning Approach
Abdulmalik Alwarafy, Mohamed Hayajneh, Mustafa Al Juboori +4
arXiv27 Oct 2026PreprintCC BY 4.0Full textGeneral mathematicsarXiv
Deep Reinforcement Learning for Blind Passive Beamforming in RIS-Assisted MIMO Systems
Oumaïma Bounhar, Thomas Chêne, Ghaya Rekaya-Ben Othman +1
arXiv29 Oct 2026PreprintCC BY-NC-SA 4.0Full textStatisticsarXiv
Reinforcement Learning for Global-Local Control of Dynamic Networks with Applications to Systemic Risk
Nils Detering, Florian Grell, Thilo Meyer-Brandis
arXiv21 Oct 2026PreprintarXiv licenceFull textOptimisation & controlarXiv

Cite

BibTeX RIS

BibTeX

@misc{lu2026neuromas,
  title = {{NeuroMAS: Multi-Agent Systems as Neural Networks with Joint Reinforcement Learning}},
  author = {Haoran Lu and Luyang Fang and Wenxuan Zhong and Ping Ma},
  howpublished = {arXiv},
  year = {2026},
  month = oct,
  doi = {10.48550/arxiv.2605.16757},
  eprint = {2605.16757},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2605.16757},
}

RIS

TY  - UNPB
TI  - NeuroMAS: Multi-Agent Systems as Neural Networks with Joint Reinforcement Learning
AU  - Haoran Lu
AU  - Luyang Fang
AU  - Wenxuan Zhong
AU  - Ping Ma
PY  - 2026
DA  - 2026/10/23
PB  - arXiv
DO  - 10.48550/arxiv.2605.16757
UR  - https://arxiv.org/abs/2605.16757
AB  - Multi-agent language systems are often built as hand-designed workflows, where agents are assigned semantic roles and communication protocols are specified in advance. We propose NeuroMAS, a method that first treats a multi-agent language system as a trainable and scalable neural-network-like architecture with LLM agents as nodes and intermediate textual signals as edges. In NeuroMAS, agent nodes are role-free but structure-aware: the topology only determines how information can flow in general, while reinforcement learning training determines how nodes communicate, specialize, and coordinate. This formulation shifts multi-agent design from workflow engineering toward architecture design, where depth, width, connectivity, and growth protocol become scalable sources of capability. Further, we provide a theoretical perspective showing why such modular textual computation is more parameter-efficient when tasks admit hierarchical decompositions. Experiments show that NeuroMAS improves significantly over both inference-time and trained multi-agent baselines. We further find that organizational scaling is path-dependent: larger systems can be challenging to train from scratch, but become feasible when grown progressively from smaller trained systems. These results suggest that learned neural multi-agent systems are a promising scaling axis for LLMs.
ER  -

Abstract

Related

Cite