ATOM: Instantiating Budget-Controllable Multi-Agent Collaboration via Nucleus-Electron Hierarchy

2026-05-25

AI Summary

This work addresses the trade-off between stability and scalability in collaborative topologies of large language model–based multi-agent systems, as well as the mismatch between computational resources and task complexity. To this end, the authors propose ATOM, a framework inspired by atomic structure that establishes a nucleus–electron hierarchical mechanism: a stable collaboration backbone (nucleus) is learned offline, while conditional agents (electrons) are dynamically activated at inference time based on input queries. Integrating task-driven reinforcement learning with query difficulty estimation, ATOM enables complexity-aware budget allocation, instantiating agents on demand to align resource consumption with task requirements. Evaluated across six diverse benchmarks, ATOM achieves state-of-the-art performance, improving token efficiency by up to 30% over strong baselines.

Abstract

Large Language Model (LLM)-based multi-agent systems rely on optimized collaboration topologies to balance performance and communication costs. However, current methods struggle with the inherent stability-extensibility trade-off and often misalign computational budgets with query difficulty. We propose \textsc{ATOM}, an adaptive framework that generates budget-controllable collaboration graphs via a novel task-driven reinforcement learning paradigm. Inspired by atomic structures, \textsc{ATOM} employs a nucleus-electron hierarchy: it maintains a stable, offline-learned collaboration backbone (the nucleus) while dynamically activating query-conditioned agents (electrons) during inference. Crucially, a complexity-aware budgeting strategy aligns resource consumption with task demands by estimating query difficulty to strictly regulate electron instantiation. Extensive experiments across six diverse benchmarks demonstrate that \textsc{ATOM} achieves state-of-the-art performance while improving token efficiency by up to $30\%$ compared to strong baselines.

Problem

• multi-agent systems
• collaboration topology
• stability-extensibility trade-off
• computational budget
• query difficulty

Innovation

• nucleus-electron hierarchy
• budget-controllable collaboration
• task-driven reinforcement learning
• complexity-aware budgeting
• multi-agent LLM systems