摘要
深度聚类在高维较大数据集中应用广泛,得益于神经网络强大的数据特征提取能力,但目前的深度聚类特征提取一般集中在神经网络的中间层,忽略了浅层特征的有用信息.为解决上述问题,提出一种基于神经网络多层特征提取的集成聚类算法(Deep Ensemble Clustering Based on Multi-Level Features,DCMLF),使用三个只有卷积层数不同而其他参数相同的网络结构提取同一个输入的不同层次特征,并进行集成聚类.通过不同层次特征组合实验验证浅层特征对聚类结果的影响,并证明该算法同经典的传统聚类算法以及经典的深度聚类算法相比,聚类性能有所提升.
Deep clustering is widely used in high-dimensional large datasets,due to the powerful capability of data feature extraction by neural networks.However,the current deep clustering feature extraction is generally concentrated in the middle layer of the neural network,ignoring the useful information of the shallow features.And different framework of neural networks are often required when facing different structured datasets.In order to solve the above problems,we propose an ensemble clustering algorithm based on multi-level feature extraction(DCMLF).Three network structures with different layers of convolutional are used to extract different hierarchical features of the input and perform integrated clustering.Experiments with different levels of feature combination prove the impact of shallow features on the clustering results,and the algorithm has improved clustering performance compared with classic traditional clustering algorithms and classic deep clustering algorithms.
作者
杜淑颖
侯海薇
丁世飞
Shuying Du;Haiwei Hou;Shifei Ding()(School of Computer Science and Technology,China University of Mining and Technology,Xuzhou,221116,China;School of Information Management,Xuzhou Vocational College of Bioengineering,Xuzhou,221000,China)
出处
《南京大学学报（自然科学版）》
CAS
CSCD
北大核心
2021年第4期575-581,共7页
Journal of Nanjing University（Natural Science）
基金
国家自然科学基金(61976216)
江苏高校“青蓝工程”。
关键词
神经网络
特征提取
深度聚类
集成聚类
neural network
feature extraction
deep clustering
ensemble clustering
