面向主动学习的模糊核聚类采样算法

发布时间：2018-03-25 14:38

本文选题：高斯核函数　切入点：聚类分析　出处：《计算机应用研究》2017年12期

【摘要】：针对主动学习中构造初始分类器难以选取代表性样本的问题,提出一种模糊核聚类采样算法。该算法首先通过聚类分析技术将样本集划分,然后分别在类簇中心和类簇边界区域选取样本进行标注,最后依此构造初始分类器。在该算法中,通过高斯核函数将原始样本空间中的点非线性变换到高维特征空间,以达到线性可聚的目的,并引入了一种基于局部密度的初始聚类中心选择方法,从而改善聚类效果。为了提高采样质量,结合划分后各类簇的样本个数设计了一种采样比例分配策略;同时,在采样结束阶段设计了一种后补采样策略,以确保采样个数达标。实验结果分析表明,所提算法可以有效地减少构造初始分类器所需的人工标注负担,并取得了较高的分类正确率。
[Abstract]:In order to solve the problem that it is difficult to select representative samples by constructing initial classifiers in active learning, a fuzzy kernel clustering sampling algorithm is proposed. Then the samples are selected from the cluster center and the cluster boundary area to annotate, and then the initial classifier is constructed. In this algorithm, the point nonlinearity in the original sample space is transformed into the high dimensional feature space by Gao Si kernel function. In order to achieve linear convergence, an initial clustering center selection method based on local density is introduced to improve the clustering effect. A sampling proportional allocation strategy is designed in combination with the sample number of each cluster, and a supplementary sampling strategy is designed at the end of the sampling stage to ensure that the sampling number is up to the standard. The experimental results show that, The proposed algorithm can effectively reduce the manual labeling burden needed to construct the initial classifier and obtain a higher classification accuracy.
【作者单位】：大连海事大学交通运输管理学院;
【基金】：国家自然科学基金资助项目(71271034) 辽宁省自然科学基金资助项目(2014025015) 青年骨干教师基金资助项目(3132016045)
【分类号】：TP311.13

【相似文献】