主动学习在标记传递算法中的应用

Application of Active Learning in Label Propagation Algorithm

为了提升标记传递算法的分类精度,提出了结合主动学习的标记传递算法。在标记传递算法中,通过主动学习挑选能最大程度提升分类性能的未标记数据给领域专家标记。在主动学习过程中,依据估计风险最小化原则挑选待标记样本。在UCI数据集上的实验表明:结合了主动学习的标记传递算法,可以减少对已标记样本数量的要求,加快学习速度和改善学习质量。

In order to enhance the classification accuracy,label propagation algorithm combined with active learning is proposed. In the label propagation algorithm,the unlabeled samples which can maximize the classification performance are selected by active learning, and the labels for selected samples are given by'domain expert'. In the process of active learning,the sample to be labeled is selected based on the estimated risk minimization principle. Experimental results on UCI data sets show that label propagation algorithm combined with active learning can reduce the amount of labeled data,speed up the learning rate,and improve the quality of learning.