摘要
大规模多视图聚类旨在解决传统多视图聚类算法中计算速度慢、空间复杂度高,以致无法扩展到大规模数据的问题.其中,基于锚点的多视图聚类方法通过使用整体数据集合的锚点集构建后者对于前者的重构矩阵,利用重构矩阵进行聚类,有效地降低了算法的时间和空间复杂度.然而,现有的方法忽视了锚点之间的差异,均等地看待所有锚点,导致聚类结果受到低质量锚点的限制.为定位更具有判别性的锚点,加强高质量锚点对聚类的影响,提出一种基于加权锚点的大规模多视图聚类算法(Multi-view clustering with weighted anchors,MVC-WA).通过引入自适应锚点加权机制,所提方法在统一框架下确定锚点的权重,进行锚图的构建.同时,为增加锚点的多样性,根据锚点之间的相似度进一步调整锚点的权重.在9个基准数据集上与现有最先进的大规模多视图聚类算法的对比实验结果验证了所提方法的高效性与有效性.
Abstract
Large-scale multi-view clustering aims to solve the problem that traditional methods cannot scale to large-scale data due to slow computational speed and high complexity.Among them,the anchor-based multi-view clustering method constructs a reconstruction matrix for the entire dataset by utilizing a set of anchor points.Clus-tering with the reconstruction matrix effectively reduces the time and space complexity of the algorithm.However,existing methods ignore the differences among anchor points and treat them equally,resulting in clustering results limited by low-quality anchor points.In order to identify more discriminative anchor points and enhance the influ-ence of high-quality anchors on clustering,a large-scale multi-view clustering algorithm based on weighted anchors(MVC-WA)was proposed.By introducing an adaptive anchor weighting mechanism,the proposed method determ-ine the weights of anchors in a unified framework for the construction of anchor graphs.Meanwhile,in order to in-crease the diversity among anchors,the weights of anchors were further adjusted according to the similarity between them.Experimental results comparing with existing state-of-the-art large-scale multi-view clustering al-gorithms on nine benchmark datasets validate the efficiency and effectiveness of the proposed method.
基金项目
国家自然科学基金(61922088)
国家自然科学基金(62006236)
国家自然科学基金(62006237)
国防科技大学科研计划项目(ZK21-23)
国防科技大学科研计划项目(ZK20-10)
高性能计算国家重点实验室自主课题(202101-15)
国家科技期刊平台
NSTL
万方数据