Multi-view unsupervised feature selection based on consensus indicator graph learning with adaptive similarity

Multi-view unsupervised feature selection based on consensus indicator graph learning with adaptive similarity

연구 분야: Artificial Intelligence

학회: Cluster Computing

Obtaining discriminating features with consensus and complementary information between different views is crucial for multi-view feature selection. However, most of the existing methods often emphasize the consistency of multi-view structures and ignore their heterogeneous information. To tackle this issue, this study proposes multi-view unsupervised feature selection of consensus indicator graph learning with adaptive similarity (CIAS). Adaptive similarity learning of multiple views plays an important role in maintaining heterogeneous information in the selected features. Concretely, structural graphs with only 0–1 values, can be learned from the dissimilarity between the samples. Then, the indicator graphs of different views are fused to obtain a shared indicator graph with a consensus structure. Finally, the feature selection matrices of different views are guided by the shared indicator graph. It can solve the proposed model efficiently using an alternating iterative approach. Furthermore, the experimental results demonstrate our algorithm’s superiority when compared to five new related algorithms.