摘要
图学习是一种广泛应用于多视图聚类的技术,它可以从多视图中学习出统一的相似图.现有的图学习方法大多只能发掘多视图的一致性,忽视了不一致的信息,这使得它们可能在学习过程中丢失视图独有的信息.为了解决这个问题,本文提出了一种融合一致性和不一致性,面向图的多视图低秩聚类框架.该方法首先将多视图分解为一致性和不一致性两个部分,然后利用自适应加权融合多视图的一致性图,并在此过程中防止权重出现平凡解;进而,提出了一种新颖的低秩融合策略,用一个统一的目标函数融合多视图一致性和不一致性,并通过谱聚类获得结果.本文还设计并实现了一种迭代优化方法来求解目标函数.最后,7个多视图数据集的对比实验验证了该方法的有效性.
Graph learning is a technique that is widely used in multi-view clustering,and it can learn a unified similar graph from multi-view.Existing graph learning methods mostly discover the consistency of multi-view,yet usually neglect the inconsistency,which makes them possible lose view-specific information during the learning process.To overcome this problem,a new graph-oriented,multi-view low-rank clustering framework that fuses consistency and inconsistency is proposed.It firstly decomposes multi-view into consistent and inconsistent parts,subsequently fuses the multi-views consistent graphs with adaptive weighting and prevents trivial weighting solutions.Further,a novel low-rank fusion strategy is proposed to fuse multi-view consistency and inconsistency with a unified objective function and it obtains results through spectral clustering.In addition,an alternate iterative optimization is designed to solve the objective function.Finally,the effectiveness of the method is verified through comparative experiments on seven multi-view datasets.
作者
滕少华
盛文涛
滕璐瑶
张巍
曾莹
TENG Shaohua;SHENG Wentao;TENG Luyao;ZHANG Wei;ZENG Ying(School of Computer,Guangdong University of Technology,Guangzhou 510006,China;School of Information Engineering,Guangzhou Panyu Polytechnic,Guangzhou 511483,China)
出处
《小型微型计算机系统》
北大核心
2025年第2期381-388,共8页
Journal of Chinese Computer Systems
基金
国家自然科学基金项目(61972102)资助。
关键词
多视图聚类
融合多视图一致性和不一致性
多视图不一致性
低秩表示
multi-view clustering
fusion of multi-view consistency and inconsistency
multi-view inconsistency
low-rank representation
