A Nonlinear Clustering Algorithm via Kernel Function and Locality Structure Learning
Linjun Chen, Guangquan Lu, Guoqiu Wen, Jiaye Li, Malong Tan · 2019
Clustering is an important task in machine learning, which is widely used in some applications such as medical image and pattern recognition. However, normal clustering algorithm have the following disadvantages: 1) some proposed clustering algorithms only consider the linear relationship of data. 2)some methods consider the global similarity of features and overlook the local similarity of features. We motivate these and propose nonlinear clustering algorithm via kernel function and locality structure learning. Firstly, our method uses the gaussian kernel function to achieve high dimensional projection so as to make the original data linearly separable.Then our method establishes the similarity matrix of data features in low-dimensional space to conduct local structure learning, which avoids the divergence of sample sets, retains the original nearest-neighbor structure relations, and improves the performance of clustering. Moreover, our method apply the sparse learning to remove the redundant features, which can improve the robustness of the model in the process of learning. Experimental results showed that our proposed method was superior to the normal linear clustering methods.