Metric Multi-View Graph Clustering
DOI:
https://doi.org/10.1609/aaai.v37i8.26188Keywords:
ML: Clustering, ML: Multi-Instance/Multi-View LearningAbstract
Graph-based methods have hitherto been used to pursue the coherent patterns of data due to its ease of implementation and efficiency. These methods have been increasingly applied in multi-view learning and achieved promising performance in various clustering tasks. However, despite their noticeable empirical success, existing graph-based multi-view clustering methods may still suffer the suboptimal solution considering that multi-view data can be very complicated in raw feature space. Moreover, existing methods usually adopt the similarity metric by an ad hoc approach, which largely simplifies the relationship among real-world data and results in an inaccurate output. To address these issues, we propose to seamlessly integrates metric learning and graph learning for multi-view clustering. Specifically, we employ a useful metric to depict the inherent structure with linearity-aware of affinity graph representation learned based on the self-expressiveness property. Furthermore, instead of directly utilizing the raw features, we prefer to recover a smooth representation such that the geometric structure of the original data can be retained. We model the above concerns into a unified learning framework, and hence complements each learning subtask in a mutual reinforcement manner. The empirical studies corroborate our theoretical findings, and demonstrate that the proposed method is able to boost the multi-view clustering performance.Downloads
Published
2023-06-26
How to Cite
Tan, Y., Liu, Y., Wu, H., Lv, J., & Huang, S. (2023). Metric Multi-View Graph Clustering. Proceedings of the AAAI Conference on Artificial Intelligence, 37(8), 9962-9970. https://doi.org/10.1609/aaai.v37i8.26188
Issue
Section
AAAI Technical Track on Machine Learning III