Abstract:

Subspace clustering is a hot research topic in machine learning. It aims at clustering data according to its potential subspace. Inspired by the collaborative training algorithm in multi-view learning, a subspace clustering algorithm induced by adaptive graph learning is proposed. The single-view data is multi-viewed by two different ways. Regularization term of the graph is iteratively updated according to the information of different views. Block diagonal affinity matrix is obtained which better reflect the clustering performance, so as to describe the data clustering results more accurately. Compared with other clustering algorithms on four standard datasets, the experimental results show the better clustering performance of the proposed method.

Key words: collaborative training, spectral clustering, subspace clustering, affinity matrix

摘要：

子空间聚类是机器学习领域的热门研究课题。它根据数据的潜在子空间对数据进行聚类。受多视图学习中协同训练算法的启发，提出一个自适应图学习诱导的子空间聚类算法，该算法首先将单视图数据多视图化，再利用不同视图的信息迭代更新图正则化项，得到更能反映聚类性能的块对角关联矩阵，从而更准确地描述数据聚类结果。在四个标准数据集上与其他聚类算法进行对比实验，实验结果显示该方法具有更好的聚类性能。

关键词: 协同训练, 谱聚类, 子空间聚类, 关联矩阵