Understanding the Information Propagation Effects of Communication Topologies in LLM-based Multi-Agent Systems

2025-05-29

AI Summary

This study investigates the causal mechanisms by which communication topology influences information propagation and collaborative efficacy in large language model (LLM)-based multi-agent systems. Addressing the fundamental trade-off between error suppression and beneficial information diffusion in sparse versus dense topologies, we propose the first causal analysis framework to quantify the differential causal effects of topological sparsity on correct versus incorrect information propagation. We introduce EIB-learner, a novel method that dynamically optimizes agent interaction graphs by adaptively fusing dense structures (to promote consensus) and sparse structures (to inhibit erroneous cascades). Our approach integrates causal inference modeling, graph neural networks, and multi-agent collaborative simulation. Experiments demonstrate an average 12.7% improvement in task performance, a 38% reduction in communication overhead, and significantly enhanced robustness against LLM-generated erroneous outputs.

Technology Category

• Multiagent Systems: Agent Communication
• Machine Learning: Causal Learning
• Cognitive Modeling & Cognitive Systems: Agent Architectures

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems

Abstract

The communication topology in large language model-based multi-agent systems fundamentally governs inter-agent collaboration patterns, critically shaping both the efficiency and effectiveness of collective decision-making. While recent studies for communication topology automated design tend to construct sparse structures for efficiency, they often overlook why and when sparse and dense topologies help or hinder collaboration. In this paper, we present a causal framework to analyze how agent outputs, whether correct or erroneous, propagate under topologies with varying sparsity. Our empirical studies reveal that moderately sparse topologies, which effectively suppress error propagation while preserving beneficial information diffusion, typically achieve optimal task performance. Guided by this insight, we propose a novel topology design approach, EIB-leanrner, that balances error suppression and beneficial information propagation by fusing connectivity patterns from both dense and sparse graphs. Extensive experiments show the superior effectiveness, communication cost, and robustness of EIB-leanrner.

Problem

• Analyzes how communication topologies affect information propagation in multi-agent systems
• Investigates why sparse and dense topologies impact collaboration efficiency differently
• Proposes a topology design balancing error suppression and information diffusion

Innovation

• Causal framework analyzes output propagation in topologies
• Moderately sparse topologies optimize task performance
• EIB-leanrner balances error and information propagation