E2-Explainer: Causal Inference for Interpretable Communication Topologies in LLM Multi-Agent Systems
The E2-Explainer framework has been developed to enhance the interpretability of communication topologies within multi-agent systems (MAS) that utilize large language models (LLMs). Current methods for generating topologies depend on black-box optimization focused solely on task-level rewards, which fails to clarify the reasons behind specific communication edge selections. E2-Explainer tackles this issue by framing topology explanation as a causal attribution challenge, pinpointing compact communication subgraphs backed by edge-level evidence of task preservation. To gather this evidence, the framework employs a Granger-style objective, enabling interpretable insights for various topology generators. This research can be found on arXiv with the identifier 2608.12921.
Key facts
- E2-Explainer is a model-agnostic framework for explaining communication topologies in LLM-based multi-agent systems.
- It formulates topology explanation as a causal attribution problem.
- It identifies compact communication subgraphs supported by edge-level evidence of task preservation.
- The framework uses a Granger-style objective to obtain evidence.
- Existing topology generation methods use black-box optimization driven by task-level rewards.
- The research is available on arXiv with identifier 2608.12921.
- The goal is to provide interpretable explanations for arbitrary topology generators.
- The framework aims to identify critical communication subgraphs responsible for successful collaboration.
Entities
Institutions
- arXiv