一种基于中心化核对齐的多核学习算法

Multiple Kernel Learning Based on Centered Kernel Alignment

核选择是核方法研究的关键内容,多核学习利用多个基核的组合代替单个核,将核选择问题转化为对组合系数的选择,有效地改进了核方法.提出一种基于中心化核对齐的二阶段多核学习方法,与传统的一阶段多核学习方法相比,该算法不仅求解效率较高,而且具有清晰的统计解释.理论分析表明,该算法不仅能显式地最大化基核的组合与输出标签之间的依赖关系,而且能隐式地最小化基核之间的冗余关系.UCI公用数据集上的实验结果表明,与经典的多核学习方法相比,该算法能显著地提高分类准确率.

Kernel selection is the key issue in kernel methods.Multiple kernel learning(MKL)utilizes a combination of multiple base kernels instead of a single one,which transforms the problem of kernel selection into the choice of combination coefficients,effectively improving kernel methods.This paper presents a two-stage MKL method based on the centered kernel alignment.Compared with the traditional MKL approaches,the proposed method not only can be efficiently computed,but also has a clear statistical interpretation.Theoretical analysis shows that the proposed method can not only explicitly maximize the dependence between the combination of base kernels and output labels,but also implicitly minimize the redundance among base kernels.The results of the UCI datasets demonstrate that the proposed method can significantly achieve better classification accuracies compared with the classical MKL approaches.