一种改进的支持向量机学习算法

分析支持向量机的几种常用的训练方法,在这个基础上提出一种改进的支持向量机学习方法。该方法将违反KKT条件程度最厉害的样本提取出来,然后缓存这些样本,作为工作集的选择范围,而且根据训练时缓存的特点,在缓存的替换上给出一种新的方法。该方法提高核缓存的命中率,减少工作集选择的代价,从而减少训练时间。实验表明,该方法能够很好地提高支持向量机的训练速度。

An Improved SVM Learning Algorithm

Analyzes several commonly used SVM training methods,and on this basis,proposes an improved support vector machine learning.The method is based on KKT conditions for the samples with the most severe violation of the extract,and then cache these samples,the range of options as the working set,and improved support vector machine shutdown conditions.The method improves the cache hit ratio to reduce the working set selection costs,thereby reducing the training time.Experiment shows that this method can improve the training speed of SVM.