Deep Multiple Auto-Encoder-Based Multi-view Clustering

Abstract Multi-view clustering (MVC), which aims to explore the underlying structure of data by leveraging heterogeneous information different views, has brought along a growth attention. algorithms based on theories have been proposed and extended in various applications. However, most existing MVC are shallow models, learn multi-view mapping low-dimensional representation space directly, ignoring nonlinear hidden each view, thus, performance is weakened certain extent. In this paper, we propose deep algorithm multiple auto-encoder, termed MVC-MAE, cluster data. MVC-MAE adopts auto-encoder capture view layer-wise manner incorporate local invariance within consistent as well complementary between any two views together. Besides, integrate learning into unified framework, such that tasks can be jointly optimized. Extensive experiments six real-world datasets demonstrate promising our compared with 15 baseline terms evaluation metrics.