半监督免疫克隆选择图划分方法

在聚类过程中利用一定量先验信息会显著提高聚类算法的性能。为了解决求解图谱划分方法NP难的问题并合理地利用一定量的先验信息，将成对限制信息引入到图谱划分方法中样本点的相似性测度，并在获得的相应的相似性矩阵的基础上，利用免疫克隆选择优化方法来优化图谱划分准则，提出了半监督免疫克隆选择图划分方法。USPS手写体数字集和UMIST人脸数据集识别的仿真实验证明了新方法的有效性。

Semisupervised immune clone selection graph partition algorithm

Using some prior information can significantly improve the performance of clustering algorithms. In order to solve the NP-hard graph partitioning problems and utilize some prior information, a semisupervised immune clone selec-tion graph partition algorithm the pairwise constraint information is introduced into the similarity measure in the graph partitioning algorithms, then the immune clonal selection algorithm is utilized to optimal the criterion of the graph parti-tioning based on the corresponding similarity matrix to obtain the solution. The experimental results on the USPS hand-written digit datasets and UMIST face datasets show that the novel method is effective.