Deep Incomplete Multi-View Clustering via Mining Cluster Complementarity

Incomplete multi-view clustering (IMVC) is an important unsupervised approach to group the data containing missing in some views. Previous IMVC methods suffer from following issues: (1) inaccurate imputation or padding for negatively affects performance, (2) quality of features after fusion might be interfered by low-quality views, especially imputed To avoid these issues, this work presents imputation-free and fusion-free deep framework. First, proposed method builds a embedding feature learning model each view individually. Our then nonlinearly maps complete into high-dimensional space discover linear separability. Concretely, paper provides implementation mapping as well shows mechanism mine cluster complementarity. This complementary information transformed supervised with high confidence, aiming achieve consistency incomplete data. Furthermore, we design EM-like optimization strategy alternately promote clustering. Extensive experiments on real-world datasets demonstrate that our achieves superior performance over state-of-the-art methods.