Abstract:Deep multiview clustering provides an efficient way to analyze the data consisting of multiple modalities and features. Recently, the autoencoder (AE)-based deep multiview clustering algorithms have attracted intensive attention by virtue of their rewarding capabilities of extracting inherent features. Nevertheless, most existing methods are still confronted by several problems. First, the multiview data usually contains abundant cross-view information, thus parallel performing an individual AE for each view and directly combining the extracted latent together can hardly construct an informative view-consensus feature space for clustering. Second, the intrinsic local structures of multiview data are complicated, hence simply embedding a preset graph constraint into multiview clustering models cannot guarantee expected performance. Third, current methods commonly utilize the Kullback-Leibler (KL) divergence as clustering loss and accordingly may yield appalling clusters that lack discriminate characters. To solve these issues, in this article we propose two new AE-based deep multiview clustering algorithms named AE-based deep multiview clustering model incorporating graph embedding (AG-DMC) and deep discriminative multiview clustering algorithm with adaptive graph constraint (ADG-DMC). In AG-DMC, a novel cross-view representation learning model is established delicately by performing decoding processes based on the cascaded view-specific latent to learn sound view-consensus features for inspiring clustering results. In addition, an entropy-regularized adaptive graph constraint is imposed on the obtained soft assignments of data to precisely preserve potential local structures. Furthermore, in the improved model ADG-DMC, the adversarial learning mechanism is adopted as clustering loss to strengthen the discrimination of different clusters for better performance. In the comprehensive experiments carried out on eight real-world datasets, the proposed algorithms have achieved superior performance in the comparison with other advanced multiview clustering algorithms.

Cross-view Topology Based Consistent and Complementary Information for Deep Multi-view Clustering

Joint Shared-and-Specific Information for Deep Multi-View Clustering

Cross-View Diversity Embedded Consensus Learning for Multi-View Clustering

Deep Embedded Complementary and Interactive Information for Multi-View Classification

Tensor-based Multi-View Clustering with Consistency Exploration and Diversity Regularization

Multi-View Clustering from the Perspective of Mutual Information

Deep Multi-View Clustering via View-Specific Representation and Global Graph

Cross-View Representation Learning-Based Deep Multiview Clustering With Adaptive Graph Constraint

Differentiable Bi-Sparse Multi-View Co-Clustering

Consistent and Complementary Graph Regularized Multi-view Subspace Clustering

Unpaired Multi-View Graph Clustering With Cross-View Structure Matching

Asymmetric double-winged multi-view clustering network for exploring diverse and consistent information

Efficient and Effective Deep Multi-view Subspace Clustering

DCCNMF: Deep Complementary and Consensus Non-negative Matrix Factorization for multi-view clustering

Multi-View Information-Theoretic Co-Clustering for Co-Occurrence Data

Enhancing multi-view clustering through common subspace integration by considering both global similarities and local structures

Deep Multiview Collaborative Clustering

A Novel Approach for Effective Multi-View Clustering with Information-Theoretic Perspective

Dual Contrast-Driven Deep Multi-View Clustering

Deep Incomplete Multi-View Clustering Via Mining Cluster Complementarity.