基于密度敏感距离的协同训练算法  

DENSITY-SENSITIVE SEMI-SUPERVISED CO-TRAINING

作　　者：徐飞裕[1] 徐荣聪[1] 

机构地区：[1]福州大学数学与计算机科学学院,福建福州350108

出　　处：《计算机应用与软件》2011年第9期229-231,共3页Computer Applications and Software

摘　　要：半监督算法作为一种能有效利用大量的未标签数据来改善少量的标签数据训练出来的分类器性能的算法,无论在理论上还是实践上都很有意义。提出一种基于密度敏感距离的协同训练算法,通过引入一种能有效描述数据的内在聚类分布的距离度量,来改善用组合分类器在少量数据集上得到的弱分类器,实验表明该方法是有效的。Semi-supervised algorithm,as an algorithm effectively exploiting large amounts of unlabeled data to improve the capability of classifier trained by small amount of labelled data,has its significance no matter in theory or in practice.A novel co-training algorithm based on density-sensitive distance is proposed in the paper.The algorithm introduces a distance metric which is able to depict inherent clustering distribution of data well and in this way to boost the performance of weak classifier derived by combined classifier on small amounts of dataset.The method is indicated effective through experiments.

关 键 词：密度敏感 协同训练 分类器组合 半监督学习 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]