摘要: Unsupervised feature selection (UFS) as an effective method to reduce time complexity and storage burden has been widely applied to various machine learning tasks. The selected features should model data distribution, preserve dat... 展开 Unsupervised feature selection (UFS) as an effective method to reduce time complexity and storage burden has been widely applied to various machine learning tasks. The selected features should model data distribution, preserve data reconstruction and maintain manifold structure. However, most UFS methods don't consider these three factors simultaneously. Motivated by this, we propose a novel joint dictionary learning method, which handles these three key factors simultaneously. In joint dictionary learning, an intrinsic space shared by feature space and pseudo label space is introduced, which can model cluster structure and reveal data reconstruction. To ensure the sparseness of intrinsic space, the ?_1-norm regularization is imposed on the representation coefficients matrix. The joint learning of robust sparse regression model and spectral clustering can select features that maintain data distribution and manifold structure. An efficient algorithm is designed to solve the proposed optimization problem. Experimental results on various types of benchmark datasets validate the effectiveness of our method. 收起