【摘要】 针对SVM训练样本集规模较大引发的学习速度慢、存储需求量大、泛化能力降低等问题,通过改进的样本点到类中心的方法来确定边界样本,从而大量缩减训练样本,提高训练速度。此外,针对非线性空间无法直接通过计算得到特征空间类中心的问题,提出了一种通过在特征空间中,寻找能生成最小超球的样本点来近似代替特征样本的替代策略,使得在保证分类精度的同时,提高了训练速度。更多还原

【Abstract】 SVM has practical applications in many fields for its superior performance at present. However,it has become a bottleneck to use SVM due to the problems including slow learning speed,large save requirement,and low generalization performance. Then an improvement reduction strategy was proposed based on the class-center,which significantly cut the amount of training samples and improved the training speed by determining the boundary training sample set. In addition,for the question that in the nonlinear space the center of the feature space class can not be directly obtained by calculating,an alternative strategy was proposed. In the alternative strategy,the sample spots that could generate the smallest hypersphere in the feature space which approximately substituted the center of feature samples were found. It improves the learning speed without reducing the classification accuracy.更多还原