论文部分内容阅读
提出一个多平面支持向量机算法——权向量多平面支持向量机(WMPSVM).该方法利用差代替Rayleigh商问题,从而避免广义特征值的奇异问题.与传统分类器不同,该方法无需求解具体的超平面,仅求解两个权向量.其决策是将测试样本归为距样本投影均值距离最近的所在的类.从广义支持向量机(GEPSVM)求解目的出发,该方法在保证得到与GEPSVM相当的计算效率的前提下,能较好地求解异或问题以及一些复杂异或问题.最后在人工数据集和UCI数据集上显示,该方法的性能要好于GEPSVM.
A multi-plane support vector machine (SVM) algorithm named Weight Vector Multi-Plane Support Vector Machine (WMPSVM) is proposed, which uses the difference instead of Rayleigh quotient to avoid the singularity problem of generalized eigenvalue. Unlike traditional classifiers, The solution is to classify the test sample as the nearest class to the projection average of the sample.For the purpose of solving the generalized support vector machine (GEPSVM), this method is guaranteed to be equivalent to GEPSVM , We can solve the XOR problem well and some complex XOR problems.Finally, we show that the performance of this method is better than GEPSVM on the artificial data set and UCI data set.