不完备数据的鲁棒多视角图学习及其聚类应用

现有多视角图学习方法主要建立在数据具有较好完备性的前提假设下,没有充分地考虑由于特征缺失引起的不完备数据的学习问题.针对此问题,提出一种不完备数据的多视角图学习方法.一方面,从局部视角内将数据重建和图学习放入同一框架,通过不完备数据补偿,实现从重建数据中学习视角专属的近邻关系,弥补特征缺失对数据分布的影响.另一方面,为了保持近邻图的二维结构,引入张量分析,从全局角度构造基于多视角的融合图学习约束,捕获缺失数据下视角间图结构的高阶潜在关联性.框架交替的优化数据重建、视角专属图学习和融合张量图结构学习,使其在迭代中相互促进,有效提高模型对不完备多视角数据的学习能力.将所提出的方法应用于两类不完备数据的多视角聚类实验,其结果表明所提出方法在多项性能指标和鲁棒性方面均优于当前主流的多视角聚类方法.

Robust multiview graph learning with applications to clustering for incomplete data

The existing multiview graph learning methods are mainly based on the premise that the data has good completeness, and do not fully consider the learning problem of incomplete data due to lack of features. Facing this problem, this paper proposes a multiview graph learning method of incomplete data. On one hand, data reconstruction and graph learning are put into the same framework from a local perspective. The view-specific neighbor relations can be learned from reconstruction data by incomplete data compensation, remedying the influence on data distribution due to feature missing. On the other hand, in order to maintain the two-dimensional structure of the nearest neighbor graph, tensor analysis is introduced to construct the fusion graph learning constraints based on multiple perspectives from a global perspective, and the capture the high-order potential relevance of graph structure between perspectives under missing data. This framework alternately optimizes data reconstruction, perspective specific graph learning and fusion tensor graph structure learning, so as to promote each other in iteration and effectively improve the learning ability of the model for incomplete multi perspective data. The proposed graph learning method is applied to two kinds of incomplete data spectral clustering experiments. The experimental results demonstrate that the proposed method outperforms the existing mainstream multiview clustering methods on both of evaluations and robustness.