基于子图分解的图聚类神经网络  

Graph clustering neural network based on subgraph decomposition

作　　者：邓祥 俞璐 姚昌华 朱瑾 Deng Xiang;Yu Lu;Yao Changhua;Zhu Jin(Army Engineering University of PLA,Nanjing,Jiangsu 210001,China;Nanjing University of Information Science)

机构地区：[1]陆军工程大学,江苏南京210001 [2]南京信息工程大学

出　　处：《计算机时代》2022年第5期6-10,共5页Computer Era

基　　金：国家自然科学基金面上项目(No.61971439);江苏省自然科学基金面上项目(No.BK20191329);中国博士后科学基金项目(No.2019T120987);南京信息工程大学人才启动经费(No.2020r100)。

摘　　要：聚类是机器学习的核心任务之一,其主要目的是将无标签数据中的不同簇数据进行分离。深度聚类算法使用深度神经网络联合优化聚类目标与特征提取,极大地提高了聚类性能。图聚类是深度聚类领域近两年研究的一个重要分支,其在处理图结构数据上有极大的优势。提出一种新的图聚类方案:基于子图分解的图聚网络,该模型在图自编码器的基础上通过构建多个子图,并在子图的嵌入空间中加以组稀疏约束达到最终的聚类目的。Clustering is one of the core tasks of machine learning. Its main purpose is to separate different cluster data from unlabeled data. Deep clustering algorithm uses deep neural network to jointly optimize clustering objectives and feature extraction,which greatly improves the clustering performance. Graph clustering is an important branch in the field of deep clustering in past two years. It has great advantages in processing graph structure data. In this paper, a new graph clustering scheme, graph clustering network based on subgraph decomposition, is proposed. Based on graph Auto-encoder, this model achieves the final clustering purpose by constructing multiple subgraphs and imposing group sparse constraints in the embedded space of subgraphs.

关 键 词：模式识别 图神经网络 深度学习 深度聚类 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]