自适应谱聚类算法研究

Adaptive spectral clustering algorithm

谱聚类能识别出在原空间中线性不可分的聚类,且其效果优于传统聚类算法.谱聚类要想获得好的效果必须选择一个合适的尺度参数,本文在传统谱聚类算法的基础上引入类似核选取的技巧,提出了一个能自动选取该尺度参数的自适应谱聚类算法.将该算法和现有的谱聚类参数选择算法作了比较,在人工数据集和UCI数据集上的实验表明,自适应谱聚类算法在很多情况下优于其它参数选择算法.

Spectral clustering has been used to identify clusters that are non-linearly separable in input space, and usually outperforms traditional clustering algorithms. However, the performances of spectral clustering are severely dependent on values of the scaling parameter. In this paper, an adaptive spectral clustering （ASC） algorithm was proposed based on traditional spectral clus- tering, which can choose the sealing parameter automatically by using techniques similar to kernel selection. The new algorithm was compared to existing parameter selection based spectral clustering algorithms on both synthetic and UCI data sets, and the experimental results validate the effectiveness of the proposed algorithm.