Unsupervised feature selection via local structure learning and sparse learning
Unsupervised feature selection via local structure learning and sparse learning
复制标题
通过局部结构学习和稀疏学习进行无监督特征选择
DOI：
10.1007/s11042-017-5381-7
复制
发表时间：
2018-11-01
影响因子：
3.6
通讯作者：
Zhu, Xiaofeng
中科院分区：
文献类型：
--
作者：
Lei, Cong;Zhu, Xiaofeng
特征自表示对噪声数据几乎不敏感，已成为无监督特征选择的支柱。然而，基于特征自表示的特征选择方法存在以下缺点：1)自表示系数矩阵是固定的，不能根据数据结构进行微调。2)没有考虑数据的流形结构，无法进一步提高特征选择的性能。为了解决上述问题，本文提出了一种结合特征自表示和流形学习的无监督特征选择算法。具体来说，我们首先利用特征自表示来构建模型。然后根据相似度矩阵动态调整自表示系数矩阵到最优状态。然后，我们使用低秩表示来探索数据的全局流形结构。最后，将稀疏学习与特征选择相结合。在12个数据集上的实验结果表明，该方法优于所有竞争方法。
Feature self-representation has become the backbone of unsupervised feature selection, since it is almost insensitive to noise data. However, feature selection methods based on feature self-representation have the following drawbacks: 1) The self-representation coefficient matrix is fixed and can not be fine-tuned according to the structure of data. 2) they do not consider the manifold structure of data, thus unable to further increase the performance of feature selection. To solve the above problems, this paper proposes an unsupervised feature selection algorithm that combines feature self-representation and manifold learning. Specifically, we first utilize feature self-representation to construct the model. After that, the self-representation coefficient matrix is dynamically adjusted to the optimal state based on the similarity matrix. Then, we use low-rank representation to explore the global manifold structure of the data. Finally, we combine sparse learning with feature selection. The experimental results on twelve datasets show that the proposed method outperforms all the competing methods.