Dynamic Generation of Multi-LLM Agents Communication Topologies with Graph Diffusion Models

2025-10-09

AI Summary

In multi-agent systems, the communication topology among LLM-based agents must dynamically balance task performance, communication overhead, and robustness—yet existing static or handcrafted topologies fail to adapt to varying task complexity. To address this, we propose the **Guided Topology Diffusion (GTD)** framework, which models topology generation as a multi-objective reward-driven iterative discrete graph diffusion process. GTD employs a conditional graph diffusion model coupled with a lightweight agent model, enabling gradient-free, real-time, and adaptive sparse topology evolution. Evaluated across multiple benchmark tasks, GTD reduces token consumption by 32% on average while improving performance on complex collaborative tasks by 18.7%. To our knowledge, GTD is the first method to achieve task-adaptive, efficient, and robust communication structure synthesis for LLM-based multi-agent systems.

Technology Category

• Multiagent Systems: Agent Communication
• Machine Learning: Large Multimodal Models (LMMs)
• Planning, Routing, and Scheduling: Planning with Language Models

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

The efficiency of multi-agent systems driven by large language models (LLMs) largely hinges on their communication topology. However, designing an optimal topology is a non-trivial challenge, as it requires balancing competing objectives such as task performance, communication cost, and robustness. Existing frameworks often rely on static or hand-crafted topologies, which inherently fail to adapt to diverse task requirements, leading to either excessive token consumption for simple problems or performance bottlenecks for complex ones. To address this challenge, we introduce a novel generative framework called extit{Guided Topology Diffusion (GTD)}. Inspired by conditional discrete graph diffusion models, GTD formulates topology synthesis as an iterative construction process. At each step, the generation is steered by a lightweight proxy model that predicts multi-objective rewards (e.g., accuracy, utility, cost), enabling real-time, gradient-free optimization towards task-adaptive topologies. This iterative, guided synthesis process distinguishes GTD from single-step generative frameworks, enabling it to better navigate complex design trade-offs. We validated GTD across multiple benchmarks, and experiments show that this framework can generate highly task-adaptive, sparse, and efficient communication topologies, significantly outperforming existing methods in LLM agent collaboration.

Problem

• Optimizing multi-agent communication topologies for task performance
• Balancing competing objectives like cost, robustness, and efficiency
• Overcoming limitations of static topologies with adaptive generation

Innovation

• Generates communication topologies using graph diffusion models
• Employs lightweight proxy model for multi-objective reward guidance
• Iteratively constructs task-adaptive topologies via gradient-free optimization