判别性正则化:一种新颖的分类器学习方法(英文)

提出了一种新颖的正则化方法-判别性正则化(Discriminative regularization,DR),为分类提供了一种通用的结合样本先验信息的方式.通过将先验信息引入到正则化项中,DR不但使分类器实际输出与期望输出之间的经验损失达到最小, 而且能在输出空间中同时最大化类间散性与最小化类内紧性.此外,通过将等式约束嵌入到目标函数中,DR的求解还可转化为解线性方程组问题,从而得到全局解析解.分类实验验证了DR的优越性.

DISCRIMINATIVE REGULARIZATION:A NEW CLASSIFIER LEARNING METHOD

A novel regularization method - discriminative regularization (DR)is presented. The method provides a general way to incorporate the prior knowledge for the classification. By introducing the prior information into the regularization term, DR is used to minimize the empirical loss between the desired and actual outputs, as well as maximize the inter-class separability and minimize the intra-class compactness in the output space simultaneously. Furthermore, by embedding equality constraints in the formulation, the solution of DR can solve a set of linear equations. Classification experiments show the superiority of the proposed DR.