摘要
针对传统的聚类算法在样本数据量不足或样本受到污染情况下的聚类性能下降问题,在经典的极大熵聚类算法(MEKTFCA)的基础上,提出了一种新的融合历史聚类中心点和历史隶属度这两种知识的基于极大熵的知识迁移模糊聚类算法。该算法通过学习由源域总结出来的有益历史聚类中心和历史隶属度知识来指导数据量不足或受污染的目标域数据的聚类任务,从而提高了聚类性能。通过一组模拟数据集和两组真实数据集构造的迁移场景上的实验,证明了该算法的有效性。
To address the issue of clustering performance degradation when traditional clustering algorithms are applied to insufficient and/or noisy data, a maximum entropy-based knowledge transfer fuzzy clustering algorithm is proposed. This improves the classical maximum entropy clustering algorithm for target domains by leveraging two kinds of knowledge from the source domain, i . e., historical clustering centers and historical degree of membership, into the objective function proposed for clustering insufficient and/or noisy target data. The effectiveness of the proposed algorithm is demonstrated by experiments on several synthetic and two real datasets.
出处
《智能系统学报》
CSCD
北大核心
2017年第1期95-103,共9页
CAAI Transactions on Intelligent Systems
基金
国家自然科学基金项目(61272210)
江苏省杰出青年基金项目(BK20140001)
江苏省自然科学基金项目(BK20130155)
关键词
知识迁移
极大熵
聚类算法
极大熵聚类
模糊聚类
knowledge transfer
maximum entropy
fuzzy clusteringclustering algorithms
maximum entropy clustering
