【摘要】 支持向量机仅仅由支持向量所决定,因此预先抽取支持向量参与训练是非常重要的。提出了一个基于同心超球面分割的支持向量预抽取方法,并在此基础上给出了HD-SVM训练算法。首先对样本的每一类分别用一些半径足够大的同心超球面进行分割,抽取出距离最优分类面较近的边界样本,这些样本最有可能成为支持向量;然后让边界样本作为初始工作集先参与训练。实验结果表明,该文的方法可以有效地对支持向量进行预抽取,避免了训练全部样本,使得训练速度明显得到提高。更多还原

【Abstract】 Support Vector Machines are only determined by support vectors,so it is very important to pre-extract SVs when training.A method for pre-extracting support vectors based on concentric hyperspheres division is presented in this paper,and HD-SVM algorithm based on this method is presented also.By dividing every class with some concentric hyperspheres,some training points close to optimal separating hyperplane can be extracted.The radii of these hyperspheres should be big enough.Then the algorithm starts with these points as the first working set,which includes SVs likely.The experiment shows the method can effectively pre-extract support vectors,so dose not train all examples,and increases the speed of training remarkably.更多还原