摘要
研究了核参数和误差惩罚参数对支持向量机推广能力的影响方式,指出核参数主要影响数据在特征空间中的分布,误差惩罚参数在特征空间中确定经验风险水平而影响SVM的性能。指出特征空间维数和学习机器复杂度并没有直接关系,讨论了结构风险最小化原则,最后给出了支持向量机和神经网络训练方法的差别和仿真试验结果。
The influences of the kernel parameters and error penalty parameter on support vector machine(SVM)'s generalization ability are studied.The test results show that the kernel parameters mainly affect the distribution of the data in the feature space,and the penalty parameter determines the level of the experimental risk in a given feature space.The results also show that the VC dimension of the feature space has no direct relationship with the complexity of SVM,and then the meaning of the structural risk minimization rule is discussed.At the end,the differences of the training methods between artificial neural network and SVM are described.
出处
《计算机工程与应用》
CSCD
北大核心
2003年第13期36-38,共3页
Computer Engineering and Applications
基金
自然科学基金委员会资助(编号:59990472)
关键词
支持向量机
核参数
结构风险最小化原则
support vector machines,kernel parameters,the structural risk minimization
