摘要
聚类分析在机器学习中起着重要的作用,学习一个好的数据表示对于聚类算法是至关重要的。深度聚类联合优化表征学习和聚类模型,已经被广泛应用于各种聚类任务中。本文概述了深度聚类的最新进展,并对其在不同领域的应用进行了总结。首先,介绍了深度聚类的基本概念和原理,从神经网络模型角度将其划分为基于自编码器、基于变分自编码器、基于生成对抗网络、基于孪生网络/对比学习和基于图神经网络的深度聚类,分别对其进行分析和总结。然后,讨论了深度聚类在图像、文本和识别检测等领域的应用。最后,对当前深度聚类研究的热点和未来发展方向进行了展望。
Clustering plays an important role in machine learning,and learning a good representation of data is essential for clustering.Deep clustering joint optimization representation learning and clustering models have been widely used in various clustering tasks.The paper overviews the recent advances in deep clustering and summarize its applications in different fields.Firstly,the basic concepts and principles of deep clustering are introduced,and from the perspective of neural network model,it is divided into autoencoder-based,variationalautoencoder-based,generative adversarial network-based,twin-based network/contrastive learning,and deep clustering based on graph neural network,which are analyzed and summarized respectively.Then,the applications of deep clustering in the fields of image,text,and recognition detection are discussed.Finally,the current hotspots and future development directions of deep clustering research are prospected.
作者
黄雪艳
张翠红
赵薇
王继奎
HUANG Xueyan;ZHANG Cuihong;ZHAO Wei;WANG Jikui(School of Information Engineering and Artificial Intelligence,Lanzhou University of Finance and Economics,Lanzhou 730000,China)
出处
《四川轻化工大学学报（自然科学版）》
CAS
2024年第4期47-57,共11页
Journal of Sichuan University of Science & Engineering(Natural Science Edition)
基金
国家自然科学基金项目(12201267)。
关键词
深度学习
神经网络
特征提取
聚类
deep clustering
neural networks
feature extraction
clustering
