摘要
分析了增量学习过程中支持向量和非支持向量的转化情况。在此基础上提出一种误分点回溯SVM增量算法,该算法先找出新增样本中被误分的样本,然后在原样本集寻找距误分点最近的样本作为训练集的一部分,重新构建分类器,这样能有效保留样本的分类信息。实验结果表明:该算法比传统的支持向量机增量算法有更高的分类精度。
The transformation between support vectors and normal vectors during an incremental learning process is analyzed. Then, a new algorithm is proposed where, a back searching operation is executed according to the error predicted samples, and the nearest vectors to the error predicted samples in original dataset is added into the training set to reconstruct the classifier. Therefore, more useful information is preserved and the classifier＇s performance is improved. The experimental results show that this algorithm works better than the traditional SVM incremental algorithm.
出处
《华东理工大学学报（自然科学版）》
CAS
CSCD
北大核心
2006年第8期989-991,共3页
Journal of East China University of Science and Technology
关键词
统计学习
支持向量机
分类
增量学习
KKT条件
statistical learning
SVM
classification
incremental learning
KKT condition
