Know the Shape, Find the Fault: Topology-Conditioned Diagnosis of Multi-Agent LLM Failures
We analyze the relationship between communication topology and failure patterns and introduce MAScope, a two-stage framework for topology-conditioned diagnosis.
ProofPaper ↗
Key points
- Multi-agent LLM systems coordinate task execution through exchanges of information among agents.
- Communication topology captures how agents exchange information and provides structural cues for distinguishing coordination failure modes.
- Using these cues for diagnosis requires establishing how topology relates to failure patterns and recovering the relevant structure from execution traces that lack explicit topology labels.
- These results show that topology-conditioned context improves failure diagnosis and supports lower-cost deployment.
Sources (1)
- [1]Know the Shape, Find the Fault: Topology-Conditioned Diagnosis of Multi-Agent LLM FailuresarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 7, 02:07 PM
We analyze the relationship between communication topology and failure patterns and introduce MAScope, a two-stage framework for topology-conditioned diagnosis.
Multi-agent LLM systems coordinate task execution through exchanges of information among agents.
Extractive summary: sentences quoted from the sources.