摘要
深度聚类由于其在数据挖掘和计算机视觉领域中处理高维数据的显著效果,已经成为一种流行的无监督学习方法.高维空间中的数据更容易存在模糊性,然而深度聚类无法直接处理数据中的模糊性.在许多实际问题中,数据之间相似性和关联性通常更集中的表现在局部邻域内,但是传统的深度聚类方法忽略了数据之间的局部关系.为了解决上述问题,本文提出了一种基于证据信息粒化的深度三支FCM聚类方法.首先,本文提出一种新的对比深度FCM聚类网络框架,将数据从复杂的原始数据空间映射到合适的深度特征空间中.其次,基于三支决策的思想,将第一阶段的聚类结果划分为正域和边界域,以便处理数据中的不确定性.最后,引入半球邻域粒化方法,为不确定样本构造信息粒.基于此,本文利用证据理论对信息粒中的信任度进行融合,从而实现对不确定数据的再分配.本文所提方法更多地关注数据的局部结构,以准确地捕捉数据的内在特征.实验结果表明,本文所提出的方法有效地提升了聚类效果.
Deep clustering has become a popular unsupervised learning method due to its remarkable effect in processing high-dimensional data in the field of data mining and computer vision.Data in high-dimensional space are more likely to have ambiguity,but deep clustering cannot directly handle the ambiguity in data.In many practical problems,the similarities and correlations between data are usually more concentrated in local neighborhoods,but traditional deep clustering methods ignore the local relationships between data.In order to solve the above problems,this paper proposes a deep three-way FCM clustering method based on evidential information granulation.First,this paper proposes a contrastive deep FCM clustering network framework to map data from complex original data space to a suitable deep feature space.Secondly,based on the idea of three-way decision,the clustering results of the first stage are divided into the positive regions and boundary regions in order to deal with the uncertainty in the data.Finally,the semiball neighborhood granulation method is introduced to construct information granules for uncertain samples.Based on this,this paper uses evidential theory to integrate trust in information particles to achieve the redistribution of uncertain data.The method proposed in this paper pays more attention to the local structure of the data to accurately capture the intrinsic characteristics of the data.Experimental results show that the method proposed in this paper effectively improves the clustering effect.
作者
郭静
蔡超越
陆杨
成晓天
樊晓雪
鞠恒荣
丁卫平
Guo Jing;Cai Chaoyue;Lu Yang;Cheng Xiaotian;Fan Xiaoxue;Ju Hengrong;Ding Weiping(School of Artificial Intelligence and Computer Science,Nantong University,Nantong 226019,China;State Key Lab.for Novel Software Technology,Nanjing University,Nanjing 210000,China)
出处
《南京师大学报(自然科学版)》
北大核心
2025年第4期106-117,共12页
Journal of Nanjing Normal University(Natural Science Edition)
基金
国家自然科学基金资助项目(62006128、U2433216、62066049)
国家重点研发计划资助项目(2024YFE0202700)
江苏高校‘青蓝工程’资助
南京大学计算机软件新技术国家重点实验室资助项目(KFKT2024B30)
江苏省自然科学基金资助项目(BK20231337)
江苏省高等学校自然科学研究重大资助项目(21KJA510004)
江苏省研究生科研与实践创新计划资助项目(SJCX24_2016、SJCX25_2001)
南通市自然科学基金青年基金资助项目(JC2024044)。
关键词
证据理论
三支决策
信息粒化
FCM聚类
深度聚类
对比学习
evidential theory
three-way decision
information granulation
FCM clustering
deep clustering
contrastive learning
