When Evidence Shapes Collaboration: Knowledge-Conditioned Topology Generation for Multi-Agent Systems
Artificial Intelligence
Summary
The authors address how groups of AI agents work together by improving the way their communication patterns are created. Current methods mostly depend on what language models already know and only use external information in a limited way, causing some confusion or missed checks in complex tasks. They introduce K-GAT, a new method that directly incorporates outside knowledge into designing how agents connect and collaborate. Tests show that K-GAT works better and uses fewer resources compared to previous approaches, especially on a challenging question-answering task.
Authors
Yangxiao Jiang, Jiarun Fan, Mingcong Xu, Yanxi Guo, Jiwen Feng, Shanqing Xu, Mengchen Qian, Wei Chen, Xiaojin Zhang
Abstract
Multi-Agent Systems (MAS) have recently moved from static workflows toward dynamically generated collaboration topologies. However, existing topology generation methods rely primarily on the parametric knowledge of large language models, with external search or retrieval used only as a reactive tool rather than an explicit determinant of collaboration structure. This leads to structure-knowledge misalignment, where systems exhibit redundant interactions or insufficient verification in knowledge-intensive tasks. We propose K-GAT (Knowledge-Guided Agent Topology Generator), a neuro-symbolic framework that formulates collaboration topology design as a knowledge-conditioned structure learning problem, integrating external evidence directly into autoregressive graph generation. Extensive experiments on knowledge-intensive benchmarks demonstrate K-GAT's efficiency and effectiveness: notably on the expert-level GPQA dataset, K-GAT outperforms the LLM-Debate baseline by a substantial margin of +15.7% in accuracy, while consuming less than half the computational tokens.