The authors propose AgentSoV, which models an LLM multi-agent system's execution as a Stream of Variation.
Modeling multi-agent LLM systems as dynamic systems for root-cause analysis
AgentSoV uses Stream of Variation to map execution trajectories to a state-space model, enabling failure attribution, influence analysis, and sensor placement across different architectures.
Big Tech
Congjing Zhang · Siyuan Xu · Yunfei Bai · Ke Ren · Xiaoning Qian · Shuai Huang
University of Washington · Texas A&M University · Amazon
Research Digest··3 min read
The authors introduce AgentSoV, a framework that treats execution of a multi-agent LLM system as a stochastic dynamic process.
§