免疫克隆选择图划分方法

为了解决谱聚类方法中大规模的相似性矩阵的存储和特征分解困难的问题,利用权核K-均值算法的目标函数和图谱划分准则的等价性,将图谱划分准则作为免疫克隆选择优化算法的亲和度函数,提出一种利用免疫克隆选择优化算法求解图谱划分问题的新方法——免疫克隆选择图划分方法。该方法在免疫克隆选择操作的过程中引入了一个个体修正算子,使得个体以更快的速度向更优的个体进化。此外,在新方法中还引入了流形距离测度来构造相似性矩阵,使得新算法可以有效处理具有复杂结构的数据。采用人工数据集、USPS手写体数字识别和UMIST人脸识别的仿真实验验证了新方法的有效性和鲁棒性。

Immune clone selection graph partition algorithm

In order to solve the problem of the storage and eigendecomposition of the similarity matrix in spectral clustering algorithms, this paper proposed a new method utilizing the immune clone selection optimizing algorithm to solve the graph partition. It utilized the equivalence of the graph partitioning and the weighted kernel K-means objectives and adopted the graph partitioning objective as the affinity function. Especially introduced an individual adjustment operator into the immune clone selection optimizing algorithm, which made the individual to evolve in better direction and higher speed. In addition, it introdced a novel distance measure to construct the similarity matrix, namely manifold distance measure, which made the method behave well in data sets with complex structure. The experimental results on six artificial datasets, the USPS handwritten digit datasets and UMIST face datasets show that the novel method is effective and robust.