Joint Multi-view unsupervised feature selection based on tensor learning

Multi-view unsupervised feature selection has gained widespread attention in the machine learning community in recent years, as it not only effectively reduces data dimensionality but also extracts key informative patterns from multi-view data. Existing studies primarily integrate graph learning with feature selection to identify the most discriminative features for learning tasks. However, these methods often rely on merging multiple graphs to construct a consensus graph or learning a global consistent graph, primarily focusing on pairwise view similarity while ignoring more complex high-order interactions among views. Moreover, the clustering results of multi-view data should have a certain consistency in different views, but this feature has not been fully utilized. To address these issues, We propose a new method called Joint Multi-view Unsupervised Feature Selection Based on Tensor Learning (JMTL). This method is based on low-rank tensors to enhance the high-order correlation and complementary information between multiple views. In addition, this method can not only model the complex interactions between views more finely, but also combine graph learning with consensus clustering to further improve the effect of feature selection. To efficiently optimize the proposed objective function, we design an optimization algorithm and conduct extensive experiments on six benchmark datasets. The experimental results demonstrate that our method outperforms state-of-the-art approaches.