KnowFlow: Empowering decision-making on networks with knowledge-streamlined agent

doi:10.1016/j.neunet.2025.108363

CSpace

	KnowFlow: Empowering decision-making on networks with knowledge-streamlined agent
	Zheng, Xiaohan 1; Wei, Lanning 2; Zhao, Huan 3; Yao, Quanming 1
	2026-04-01
发表期刊	NEURAL NETWORKS
ISSN	0893-6080
卷号	196 页码:11
摘要	Effective decision-making on networks often relies on learning from graph-structured data, where Graph Neural Networks (GNNs) play a central role. However, real-world applications differ substantially in both their tasks and graph characteristics, requiring substantial knowledge to understand the underlying problems and the structure of the data. Despite this need, existing methods still lack explicit guidelines for incorporating such knowledge into the design of effective GNNs. To mitigate this gap, we propose to prepare and apply graph learning knowledge to empower the GNN design. Specifically, we first gather diverse resources on graph learning, then develop an LLM-based agent to extract and retrieve knowledge. Subsequently, four graph learning agents are designed to leverage knowledge to enhance problem understanding, GNN design and evaluation procedures. Then, the problems can be solved by utilizing knowledge across these procedures, and the proposed method is dubbed KnowFlow. We evaluate KnowFlow on twelve datasets covering different tasks, including node classification, graph classification, and link prediction. The results show that it achieves the best performance among baselines with comparable resource costs, demonstrating both effectiveness and efficiency.
关键词	Neural architecture search Knowledge base LLM-based agents Automated machine learning
DOI	10.1016/j.neunet.2025.108363
收录类别	SCI
语种	英语
WOS研究方向	Computer Science ; Neurosciences & Neurology
WOS类目	Computer Science, Artificial Intelligence ; Neurosciences
WOS记录号	WOS:001637797300001
出版者	PERGAMON-ELSEVIER SCIENCE LTD
引用统计
文献类型	期刊论文
条目标识符	http://119.78.100.204/handle/2XEOYT63/42940
专题	中国科学院计算技术研究所
通讯作者	Yao, Quanming
作者单位	1.Tsinghua Univ, Dept Electroni Engn, Beijing, Peoples R China 2.Univ Chinese Acad Sci, Inst Comp Technol, Chinese Acad Sci, Beijing, Peoples R China 3.Noumena AI, Beijing, Peoples R China
推荐引用方式 GB/T 7714	Zheng, Xiaohan,Wei, Lanning,Zhao, Huan,et al. KnowFlow: Empowering decision-making on networks with knowledge-streamlined agent[J]. NEURAL NETWORKS,2026,196:11.
APA	Zheng, Xiaohan,Wei, Lanning,Zhao, Huan,&Yao, Quanming.(2026).KnowFlow: Empowering decision-making on networks with knowledge-streamlined agent.NEURAL NETWORKS,196,11.
MLA	Zheng, Xiaohan,et al."KnowFlow: Empowering decision-making on networks with knowledge-streamlined agent".NEURAL NETWORKS 196(2026):11.