Evaluating and Improving Graph-based Explanation Methods for Multi-Agent Coordination

2025-02-14

AI Summary

Existing graph neural network (GNN) explanation methods lack sufficient interpretability for communication channels in multi-agent collaborative decision-making. Method: We systematically evaluate the limitations of current GNN explainers in identifying critical communication links and propose an attention entropy regularization mechanism to enhance the explainability-friendliness of Graph Attention Network (GAT)-based policies. This method imposes theoretically grounded information entropy constraints to improve discriminability between the explanation subgraph and its complement. Contribution/Results: Our approach maintains high task performance (success rate ≥98.5%) while significantly improving explanation quality—achieving an average 12.7% AUC gain across three tasks and three scalability settings. It is the first work to rigorously characterize the applicability boundaries of graph explanation methods in multi-agent settings, establishing a novel paradigm and practical toolkit for explainable multi-agent learning.

Technology Category

• Multiagent Systems: Agent Communication
• Machine Learning: Graph-based Machine Learning
• Humans and AI: Explainable AI (XAI) for Human Understanding

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Agentic search

Abstract

Graph Neural Networks (GNNs), developed by the graph learning community, have been adopted and shown to be highly effective in multi-robot and multi-agent learning. Inspired by this successful cross-pollination, we investigate and characterize the suitability of existing GNN explanation methods for explaining multi-agent coordination. We find that these methods have the potential to identify the most-influential communication channels that impact the team's behavior. Informed by our initial analyses, we propose an attention entropy regularization term that renders GAT-based policies more amenable to existing graph-based explainers. Intuitively, minimizing attention entropy incentivizes agents to limit their attention to the most influential or impactful agents, thereby easing the challenge faced by the explainer. We theoretically ground this intuition by showing that minimizing attention entropy increases the disparity between the explainer-generated subgraph and its complement. Evaluations across three tasks and three team sizes i) provides insights into the effectiveness of existing explainers, and ii) demonstrates that our proposed regularization consistently improves explanation quality without sacrificing task performance.

Problem

• Improving GNN explanation methods
• Identifying influential communication channels
• Enhancing multi-agent coordination explainability

Innovation

• GNNs enhance multi-agent coordination
• Attention entropy regularization improves explanations
• Minimizing entropy focuses on influential agents