Multi-view Clustering via Multiple Auto-Encoder

Abstract

Multi-view clustering (MVC), which aims to explore the underlying structure of data by leveraging heterogeneous information of different views, has brought along a growth of attention. Multi-view clustering algorithms based on different theories have been proposed and extended in various applications. However, existing of most MVC algorithms are shallow models. They learn structure information of multi-view data by mapping multi-view data to low-dimensional representation space directly, which ignore the Non-linear structure information hidden in each view. This weakens the performance of multi-view clustering to a certain extent. In this paper, we propose a multi-view clustering algorithm based on multiple Auto-Encoder, named MVC-MAE, to cluster multi-view data. MVC-MAE algorithm adopts Auto-Encoder to capture the non-linear structure information of each view in a layer-wise manner. To exploit the consistent and complementary information contained in different views, we also incorporate the local invariance within each view and consistent and complementary information between any two views. Besides, we integrate the representation learning and clustering into a unified step, which jointly optimizes these two steps. Extensive experiments on three real-world datasets demonstrate a superior performance of our algorithm compared with 13 baseline algorithms in terms of two evaluation metrics.