基于调整学习的聚类算法

Clustering Algorithm Based on Fine-Tuned Learning

调整学习是一种逐步求精的近似学习方法,是提高局部搜索解质量的重要途径之一。该方法调用调整算子填平局部最优解陷阱,构造一系列不同粒度的搜索空间,降低局部最优解对解质量的影响。利用调整学习的基本原理设计了聚类算法框架CAT-L,并给出了适合处理聚类问题的噪声平滑调整算子。实验对比了经典FCM算法与FCM-CAT-L(以FCM算法作为CAT-L框架的聚类算子)算法的聚类质量。实验结果表明,调整学习方法对提高聚类质量是有效的。

Fine- tuned learning, one of the important ways to increase the quality of solution of local search algorithm, is an approximation learning method. A fine- tuned operator is used to create a series of different granularity search spaces in which most traps haven＇t been smoothed, so the influence of the traps is reduced. In this paper, a clustering algorithm framework CAT_L （Clustering Algorithm based on Fine - tuned Learning）, stimulated by fine- tuned learning, is proposed. Simultaneously, a noise smoothing fine - tuned oparator, which adapts to deal with clustering problem, is designed. Compared the quality of classical FCM and FCM - CAT_L through experiments, the results show that fine- tuned learning is very effective for increasing the quality of duster.