IE-Prop：估计 LLM 多智能体系统中的传播风险
Estimation of Propagated Risk in LLM-based Multi-Agent Systems
AI 导读
新提出的 Interaction-Effect Propagation（IE-Prop）方法可从多智能体执行轨迹中估计 LLM 多智能体系统的下游错误风险，对正确与错误上游智能体分别建模交互效应，并沿执行图递归传播风险估计。在三种多智能体架构和三个 benchmark 上评估，最终节点 AUROC 较既有基线平均提升约 0.05；在顺序系统中多步风险追踪带来 +0.03 AUROC 增益。
正文 · 原文
Abstract:Large Language Model (LLM) based multi-agent systems introduce a distinctive reliability challenge in which actions taken by upstream agents can affect the correctness of downstream agents. While prior work has investigated error propagation and developed methods for estimating uncertainty inherited from upstream agents, existing approaches do not explicitly model how realized sender-receiver interactions contribute to downstream error risk under different sender correctness states. In this work, we study agent-level error-risk estimation from multi-agent execution traces. We propose Interaction-Effect Propagation (IE-Prop), an interaction-conditioned model that learns separate effects for correct and incorrect upstream agents. For each sender-receiver interaction, the model combines these conditional effects using the sender's estimated error risk, and recursively propagates the resulting risk estimates through the execution graph. We evaluate our approach across three multi-agent architectures and three benchmarks. Across all evaluated architectures, our approach raises the final-node AUROC over the established baseline by approximately 0.05 points on average. Furthermore, tracking risk across multiple agent steps contributes a distinct +0.03 AUROC gain in sequential systems. These results show that realized inter-agent interactions provide predictive information beyond receiver-local evidence, while the value of multi-step risk tracking depends on the system's information-flow structure.
| Subjects: | Artificial Intelligence (cs.AI); Multiagent Systems (cs.MA) |
| Cite as: | arXiv:2610.33282 [cs.AI] |
| (or arXiv:2610.33282v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.33282 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Shreyan Ganguly [view email]
[v1]
Fri, 23 Oct 2026 20:46:33 UTC (2,179 KB)
来源：arXiv cs.AI · arxiv.org