摘要
为了解决大样本集标记工作问题和分类器对新样本分类适应能力差的问题,结合球结构支持向量机,提出了一种主动学习自适应性分类方法。该方法根据主动学习思想,以边界近邻策略迭代选取最有价值的样本,初始训练分类器,再依据增量学习方法选取包含新信息的样本,以阶段跟新方式重新训练分类器,并根据余弦相似度对内存中支持向量进行控制。实验结果表明,该方法既减少了标记开销,又保持了分类器分类性能的稳定性和延续性。
In order to deal with the problem on large samples set marked and classifier's poor adaptive ability to new samples, a research is made. A active and adaptive classifier method is proposed which combined with sphere structured support vector ma chine. The method firstly adopt active idea in which the most valuable training samples ehoosed according to close neighbor choosing strategy and the classifier trained initially, then in incremental learning idea the new samples with new information picked and classifier re-trainned according by step updated strategy, the number of support vector in memory is controlled by cosine similarity. The experiment show that the method not only reduces the marking time, but also maintain a good classification performance.
出处
《计算机工程与设计》
CSCD
北大核心
2012年第11期4116-4120,共5页
Computer Engineering and Design
关键词
主动学习
球结构支持向量机
训练样本
增量学习
支持向量
内存控制
active learning
sphere structured SVM
training samples
incremental learning
support vector
memory control