UMAS：多智能体 LLM 系统的系统级不确定性量化
原标题：UMAS: System-Level Uncertainty Quantification for Multi-Agent LLM Systems
AI 摘要
研究者提出 UMAS，一个面向多智能体 LLM 系统（MAS）的系统级不确定性量化框架。该方法无需额外参数，也不增加推理成本：它根据每次生成的内在置信度估计智能体之间影响的可信度，并据此调节各节点不确定性的更新方式，再将节点级不确定性聚合为系统级估计。在 Debate 和 DyLAN 两种多智能体架构、GPT-4o mini 与 Llama 4 Scout 两个模型以及五个基准上，UMAS 平均比最强基线高 10.54 个 AUROC 点，并可用于幻觉检测与不确定性感知的答案选择，在部分任务上把 MAS 准确率最多提升 12.6 点。
正文节选
: System-Level Uncertainty Quantification for Multi-Agent LLM Systems Abstract Multi-Agent Systems (MAS) are a promising paradigm for improving the reasoning capabilities of large language models, yet they remain prone to hallucinations and erroneous outputs. Uncertainty Quantification (UQ) is crucial for assessing system reliability, but existing methods are largely designed for single-agent settings; how to exploit MAS structure for UQ while keeping inference cost-neutral remains underexplored