说明：  经验风险由损失函数定义，描述了分类器所给出的分类结果的准确程度；结构风险由分类器参数矩阵的范数定义，描述了分类器自身的复杂程度以及稳定程度，复杂的分类器容易产生过拟合，因此是不稳定的。若一个分类器通过最小化经验风险和结构风险的线性组合以确定其模型参数：
(Empirical risk is defined by loss function, which describes the accuracy of the classification results given by the classifier. Structural risk is defined by the norm of the parameter matrix of the classifier, which describes the complexity and stability of the classifier itself. Complex classifiers are prone to over-fitting, so they are unstable. If a classifier determines its model parameters by minimizing the linear combination of empirical risk and structural risk:)