节点文献
支持向量机的若干问题的研究
【作者】 刘鹏;
【导师】 陈笑蓉;
【作者基本信息】 贵州大学 , 计算机应用技术, 2007, 硕士
【摘要】 本文提出了一种改进的支持向量分类方法和一种针对支持向量机的增量学习算法。根据支持向量机中支持向量不会出现在两类样本集间隔以外的正确划分区的理论,通过引入类质心,类半径,类质心距等概念,从而较好地解决快速而准确的删除非支持向量的问题;引入了类向心度的概念,解决了当两类样本集混淆严重的时候如何更加精确地进行剔除混淆点,保证算法泛化性的问题。实验表明,采用这种改进的算法既能快速精确地对训练样本进行删减,又可以当两类训练样本集混淆较严重时又能较好地解决泛化性问题。通过引入相似度上限、相似度上限,以及可靠度下限等概念,提出了一种可以快速收敛的支持向量机增量学习算法。实验表明,采用这种改进支持向量机增量算法能使决策函数快速地向真实的决策函数逼近。
【Abstract】 An improved SVM algorithm is proposed based on the theory that support vectors will not appear in the areas which out of the interval between two classes. Benefiting from the concepts of class-radius, class-centroid-distance and class-centripetal force. we can delete those non-SVs effectively with high accuracy and generalization capability even the data was promiscuous. The experimental results show that, comparing with other algorithms; our method achieved a satisfactory result. An improved incremental SVM algorithm is also proposed, which is based on the concepts of Comparability and dependability. The experimental results show that the improved incremental SVM algorithm can make decision-function approach the real one quickly.
【Key words】 Support vector machine; Class-centroid; Class-radius; Class-centroid-distance; Incremental learning; Comparability; Dependability;
- 【网络出版投稿人】 贵州大学 【网络出版年期】2007年 05期
- 【分类号】TP18
- 【被引频次】3
- 【下载频次】352