论文部分内容阅读
在各种支持向量机(SVM)训练算法中,比较突出的训练算法是序贯最小优化(SMO)算法。样本取样SMO算法是在不改变样本分布的前提下对原始训练集进行取样从而压缩样本数量,但由于样本取样具有随机性,如何有效缩减取样范围是改进该算法的主要方向。为此根据边界向量不一定是支持向量,但支持向量一定是边界向量这一理论,得出边界向量集是包含所有支持向量的集合,先提取边界向量再取样,把取样的范围减小到边界向量集里,缩短样本取样SMO算法的时间。实验表明,基于边界向量的样本取样SMO算法的性能要比原算法更优。
Among the various SVM training algorithms, the most prominent training algorithm is Sequential Minimization (SMO) algorithm. Sample Sampling The SMO algorithm samples the original training set to compress the sample number without changing the sample distribution. However, due to the randomness of sample sampling, how to effectively reduce the sampling range is the main direction to improve the algorithm. For this reason, the boundary vector is not necessarily a support vector, but the support vector must be a boundary vector. The result shows that the boundary vector set is a set containing all the support vectors. The boundary vectors are first extracted and resampled to reduce the sampling range to the boundary In the vector set, shorten the sample sampling time of the SMO algorithm. Experiments show that the SMO algorithm based on the boundary vector is better than the original algorithm.