Multi-view semi-supervised feature selection based on adaptive graph and tensor learning

Multi-view semi-supervised feature selection based on adaptive graph and tensor learning

연구 분야: Artificial Intelligence

학회: Applied Intelligence

Multi-view feature selection has gained popularity as a study area in the machine learning field, because it can effectively reduce the dimensionality of data. Due to the scarcity of labeled information, numerous algorithms for unsupervised feature selection have surfaced, while supervised or semi-supervised feature selection algorithms are relatively few. However, when there is a certain amount of labeled information, unsupervised learning algorithms fail to explore the true structure of samples due to their inherent limitations. A multi-view semi-supervised feature selection method based on adaptive graph and tensor learning is proposed to effectively reduce the dimension of data with few labels. The method completes the feature selection task by exploring high-order connections between views and discriminative label information. An efficient algorithm is designed to solve the resulted optimization problem. Compared with other mainstream methods, we have achieved a certain improvement in classification accuracy on multiple basic datasets which proves the superiority and universality of our method. We also conducted experiments such as convergence analysis to prove the effectiveness of our proposed method.