Abstract: Generalizing kernel consistency method, semi-supervised learning algorithm is presented based on kernel strategy. Five different measures and relation among them are deeply analyzed. Relation between arguments of different measures and performance of algorithm is experimentally studied and performance of algorithm is compared by using different measures method. Experimental results show that performance of GCM algorithm for using the exponential measure is superior to other measures and performance of GCM algorithm for using the Euclidean measure is inferior to other measures. Moreover, arguments for different measures impact on the performance of algorithm.

Key words: semi-supervised learning, kernel, measure, classification

摘要： 通过扩展核一致性方法，提出基于核策略的半监督学习算法GCM，研究5种不同度量方法中参数与算法性能的关系，对使用不同度量的GCM算法的性能进行比较。实验结果表明，使用指数度量的GCM算法的性能最优，而使用欧几里得度量的GCM算法的性能最差。不同度量中的参数取值对算法的性能具有一定的影响。

关键词: 半监督学习, 核, 度量, 分类

CLC Number: 

• TP389.1