论文部分内容阅读
训练SVM的本质是解决二次规划问题,在实际应用中,如果用于训练的样本很大,标准的二次型优化技术很难应用。最近有学者将S-K算法和核方法相结合,用L_2代价函数来解决SVM问题。这种算法解决了非线性情况和数据不可分的情况。由于它所需的记忆存储和数据量之间呈线性关系,因此这种算法可以用来解决大规模样本集的训练问题。文章对由S-K算法构造最大间隔分类器进行了研究,并用基于S-K算法的核方法构造C-SVM分类器,并取得了令人满意的效果。
The essence of training SVM is to solve the quadratic programming problem. In practice, if the samples used for training are very large, the standard quadratic optimization technique is difficult to apply. Recently, some scholars combine the S-K algorithm with the kernel method to solve the SVM problem with the L2 cost function. This algorithm solves the non-linear situation and the data can not be separated. Because of the linear relationship between the amount of memory and the amount of memory it requires, this algorithm can be used to solve the training problem of large sample sets. In this paper, the maximum interval classifier constructed by S-K algorithm is studied, and the C-SVM classifier is constructed by using the kernel method based on S-K algorithm. The satisfactory results are obtained.