随着大规模无标注数据的快速增长,有效和灵活地对其进行分析以挖掘其潜在价值具有重要现实意义也面临较大挑战。真实世界中的数据构成较为复杂，传统的划分式聚类方法在灵活性以及复杂信息融合方面难以适应实际需求。本项目拟研究多视角协同的层次化聚类，以同时提高聚类的准确性和灵活性。具体包括：1）研究多特征协同的层次聚类及特征缺失条件下的多特征协同层次聚类；2）研究特征-结构协同的层次聚类，有效利用多种特征和多种结构并存的复杂多视角信息；3）研究多视角协同维度约减方法及多视角协同采样方法，对原始数据进行维度和数量上的约减以提高算法伸缩性。研究成果可广泛应用于多视角数据层次聚类分析，对广义的特征融合、大规模聚类问题具有重要的理论价值和借鉴意义。

With the rapid growth of large scale unlabeled data, it is very important but challenging to effectively and flexibly analyze these data for the underlying value. Since the constitution of the data in real world is usually complex, traditional clustering methods cannot deal with them for lack of flexibility and the ability of integrating complex information. Therefore, for the effectiveness and flexibility in clustering analysis, in this project we propose the multi-view hierarchical clustering framework. Specifically, the proposal consists of three components: 1) We carry out research on how to efficiently make use of the multiple features, furthermore, we will improve the method for the multi-view data with uncompleted views which is ubiquitous in real applications; 2) Based on the former component, we further focus on effectively making use of multiple features as well as multiple structures simultaneously to make the multi-view hierarchical clustering framework more general for many applications; 3) For the better scalability, we will design the multi-view dimension reduction method and sampling based method. The research outputs will support the ubiquitous multi-view hierarchical clustering analysis and provide important theory to general multi-feature learning and large scale clustering.