KnowFlow: Empowering decision-making on networks with knowledge-streamlined agent
文献类型：期刊论文
| 作者 | Zheng, Xiaohan2; Wei, Lanning1; Zhao, Huan3; Yao, Quanming2 |
| 刊名 | NEURAL NETWORKS
 |
| 出版日期 | 2026-04-01 |
| 卷号 | 196页码:11 |
| 关键词 | Neural architecture search Knowledge base LLM-based agents Automated machine learning |
| ISSN号 | 0893-6080 |
| DOI | 10.1016/j.neunet.2025.108363 |
| 英文摘要 | 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. |
| 资助项目 | National Key Researchand Development Program of China[.2023YFB2903904] ; National Natural Science Foundation of China[92270106] ; Beijing Natural Science Foundation[4242039] |
| WOS研究方向 | Computer Science ; Neurosciences & Neurology |
| 语种 | 英语 |
| WOS记录号 | WOS:001637797300001 |
| 出版者 | PERGAMON-ELSEVIER SCIENCE LTD |
| 源URL | [http://119.78.100.204/handle/2XEOYT63/42939]  |
| 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.Univ Chinese Acad Sci, Inst Comp Technol, Chinese Acad Sci, Beijing, Peoples R China 2.Tsinghua Univ, Dept Electroni Engn, 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. |
入库方式： OAI收割
来源：计算技术研究所
浏览0
下载0
收藏0
其他版本
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。