论文部分内容阅读
通过对核矩阵的研究,利用核矩阵的对称正定性,采用核校准的方法提出了一种SVM最优模型选择的算法———OMSA算法.利用训练样本不通过SVM标准训练和测试过程而寻求最优的核参数和相应的最优学习模型,弥补了传统SVM在模型选择上经验性强和计算量大的不足.采用该算法在UCI标准数据集和FERET标准人脸库上进行了实验,结果表明,通过该算法找到的核参数以及相应的核矩阵是最优的,得到的SVM分类器的错误率最小.该算法为SVM最优模型选择提供了一种可行的方法,同时对其他基于核的学习方法也具有一定的参考价值.
Based on the research of nuclear matrix and using the symmetry positivity of nuclear matrix, an algorithm for selecting the optimal model of SVM (OMSA algorithm) is proposed by using nuclear calibration method. The training samples are not sought through the standard training and testing process of SVM The optimal kernel parameters and the corresponding optimal learning model make up for the shortcomings of the traditional SVM in the model selection of strong experience and large amount of computation.Using the algorithm in the UCI standard data set and FERET standard face database for experiments, The results show that the kernel parameter and the corresponding kernel matrix found by this algorithm are optimal, and the error rate of the SVM classifier is the minimum.The algorithm provides a feasible method for the optimal model selection of SVM, Nuclear learning methods also have some reference value.