Extreme Support Vector Machine Classifier
Extreme Support Vector Machine Classifier
复制标题
DOI：
10.1007/978-3-540-68125-0_21
复制
发表时间：
2008-05
期刊：
影响因子：
--
通讯作者：
Qiuge Liu;Qing He;Zhongzhi Shi
中科院分区：
文献类型：
--
作者：
Qiuge Liu;Qing He;Zhongzhi Shi
与以往的SVM算法不同，SVM算法利用核函数来计算特征空间中数据点的点积，这里的点通过单隐层前馈网络（SLFN）显式映射到特征空间，其输入权重随机生成。在理论上，这种公式，它可以被解释为一种特殊形式的正则化网络（RN），往往提供更好的泛化性能比算法SLFNs-Extreme Learning Machine（ELM），并导致一个非常简单和快速的非线性SVM算法，只需要一个潜在的小矩阵的逆，其阶数与训练数据集的大小无关。实验结果表明，所提出的极值SVM在几乎所有情况下都能产生比ELM更好的泛化性能，并且在精度相当的情况下，运行速度比其他非线性SVM算法快得多。
Instead of previous SVM algorithms that utilize a kernel to evaluate the dot products of data points in a feature space, here points are explicitly mapped into a feature space by a Single hidden Layer Feedforward Network (SLFN) with its input weights randomly generated. In theory this formulation, which can be interpreted as a special form of Regularization Network (RN), tends to provide better generalization performance than the algorithm for SLFNs—Extreme Learning Machine (ELM) and leads to a extremely simple and fast nonlinear SVM algorithm that requires only the inversion of a potentially small matrix with the order independent of the size of the training dataset. The experimental results show that the proposed Extreme SVM can produce better generalization performance than ELM almost all of the time and can run much faster than other nonlinear SVM algorithms with comparable accuracy.