AI summaryⓘ
The authors address limitations in current multi-agent systems that use large language models, where agents are not specialized enough for different parts of complex tasks. They propose MoRSE, a system that breaks tasks into smaller steps and assigns agents specific roles and subtasks, improving specialization both in task assignment and internal parameters. To efficiently adapt parameters for each role and subtask, they introduce a flexible expert module combined with a routing method to decide which expert to use. They also develop a training approach that separately optimizes these experts despite complex reward signals. Their tests on code-generation tasks show better overall and step-by-step results, and the specialized agents work well on new tasks too.
large language modelsmulti-agent systemstask decompositionparameter adaptationrole specializationLoRA (Low-Rank Adaptation)semantic routingpolicy optimizationcode generationDir acyclic graph
Authors
Peiwen Li, Shiyang Zhang, Yangtian Zhang, Sizhuang He, David van Dijk, Rex Ying
Abstract
Large language model-based multi-agent systems have recently shown strong potential for complex, long-horizon tasks. However, existing methods mainly rely on coarse prompt-level differentiation without parameter adaptation for diverse subtasks, resulting in insufficient inter-agent heterogeneity and limited specialized capability that bottleneck performance on tasks with complex requirements. To address this, we introduce a Task-Oriented Multi-Agent System with Mixture of Role-Subtask Experts (MoRSE) that distinguishes agents with (role, subtask)-conditional specialization at both the task structure and parameter levels. To make agents' responsibility explicit at the task structure level, we formulate a task-oriented multi-agent system that decomposes each task into a dependency-aware Directed Acyclic Graph of subtasks and assigns each agent a specific (role, subtask), introducing task-level specialization across collaborating agents. Additionally, to address the diverse role and subtask parameter adaptation demands, we propose a dynamic Mixture of (role, subtask) LoRA Experts module with a prototype-based semantic router for subtasks, augmenting agents with parameter-level specialization on a shared LLM substrate cost-effectively. Then, to co-optimize experts and router stably under sparse task rewards, we further propose a hierarchical group-relative policy optimization with two-layer credit assignment that isolates expert updates from the cross-route variance introduced by routing decisions, disentangling expert quality from routing quality. Experiments on code-generation benchmarks across three backbones demonstrate the effectiveness of our approach, with improvements in both whole-task and step-wise performance, and the gains from trained specialization generalize across held-out task categories and domains.