核聚类启发式转导SVM

传统转导支持向量机有效地利用了未标记样本，具有较高的分类准确率，但是计算复杂度较高。针对该不足，论文提出了一种基于核聚类的启发式转导支持向量机学习算法。首先将未标记样本利用核聚类算法进行划分，然后对划分后的每一簇样本标记为同一类别，最后根据传统的转导支持向量机算法进行新样本集合上的分类学习。所提方法通过对核聚类后同一簇未标记样本赋予同样的类别，极大地降低了传统转导支持向量机算法的计算复杂度。在MNIST手写阿拉伯数字识别数据集上的实验表明，所提算法较好地保持了传统转导支持向量机分类精度高的优势。

Heuristic Transductive Support Vector Machine Algorithm Based on Kernel Clustering

As the traditional transductive support vector machine makes full use of unlabeled samples, it has a high classification preci- sion, but high computational complexity. For this shortcoming, this paper proposes a heuristic transductive support vector machine algorithm based on kernel clustering. First, the unlabeled samples are divided by kernel clustering algorithm. Then label samples in the same cluster are marked with the same label. Finally, a classifier with traditional transductive support vector machine on the new training set is learned. By labeling samples in the same cluster with the same label, the proposed algorithm can effectively reduce the computational complexity. Ex- periment on MNIST handwritten Arabic digits recognition data sets shows that the proposed algorithm can effectively maintain the advantage of traditional transductive support vector machine on classification precision.