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

TL;DR

NeuroMAS introduces a neural-network-inspired framework for multi-agent language systems, enabling scalable, trainable, and structure-aware agent architectures that improve performance and efficiency.

Contribution

It redefines multi-agent system design as architecture optimization, leveraging reinforcement learning to enhance scalability and coordination among agents.

Findings

NeuroMAS outperforms existing inference-time and trained multi-agent baselines.

Organizational scaling is more effective when systems are grown progressively from smaller trained units.

Modular textual computation is more parameter-efficient for hierarchical tasks.

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.