RADAR: Redundancy-Aware Diffusion for Multi-Agent Communication Structure Generation
For multi-agent LLM systems, RADAR addresses the problem of fixed or single-step communication topologies that waste tokens on simple tasks and limit capability on complex tasks.
RADAR is a redundancy-aware generative framework for multi-agent communication topology that uses step-by-step graph diffusion guided by effective graph size, achieving higher accuracy, lower token consumption, and greater robustness across six benchmarks.
Compared with individual agents, large language model based multi-agent systems have shown great capabilities consistently across diverse tasks, including code generation, mathematical reasoning, and planning, etc. Despite their impressive performance, the effectiveness and robustness of these systems heavily rely on their communication topology, which is often fixed or generated in a single step. This restricts fine-grained structural exploration and flexible composition, resulting in excessive token utilization on simple tasks while limiting capability on complicated tasks. To mitigate this challenge, we introduce RADAR, a redundancy-aware and query-adaptive generative framework that actively reduce communication overhead. Motivated by recent progress in conditional discrete graph diffusion models, we formulate communication topology design as a step-by-step generation process, guided by the effective size of the graph. Comprehensive experiments on six benchmarks demonstrate that RADAR consistently outperforms recent baselines, achieving higher accuracy, lower token consumption, and greater robustness across diverse scenarios. Our code and data are available at https://github.com/cszhangzhen/RADAR.