智能系统学报2025，Vol.20Issue(2)：316-328,13.DOI:10.11992/tis.202311044
视图映射和循环一致性生成的不完整多视图聚类
Incomplete multiview clustering based on view mapping and cyclic consistency generation
摘要
Abstract
Traditional clustering assumes that each view is complete without accounting for incomplete views caused by data corruption,device failures,and other factors.To address this issue,most existing methods rely on kernel and non-negative matrix factorization without explicitly compensating for data loss in each view,and the potential representation of learning does not fully account for clustering tasks.An incomplete multiview clustering method(MG-IMC)with view mapping and cyclic consistency generation is designed to address the aforementioned limitation.This method leverages existing data information to generate missing data for each view through a single generative adversarial network,using shared potential representations provided by other views.Weighted adaptive fusion is applied to capture enhanced gen-eric structures on the generated complete dataset,followed by clustering based on KL-divergence loss.The joint train-ing of encoding common representations and generating missing data allows the model to recover missing data while simultaneously generating clustering-friendly common representations.Experiment results demonstrate that this al-gorithm outperforms existing methods in clustering performance.关键词
数据挖掘/聚类/多视图学习/不完全多视图聚类/深度学习/自动编码器/生成对抗性网络/KL散度Key words
data mining/clustering/multiview learning/incomplete multiview clustering/deep learning/autoencoder/generate adversarial networks/KL-divergence分类
信息技术与安全科学引用本文复制引用
王英博,郭凯雪..视图映射和循环一致性生成的不完整多视图聚类[J].智能系统学报,2025,20(2):316-328,13.基金项目
辽宁省教育厅基础研究项目(LN2020JCL029). （LN2020JCL029）
