摘要
核机器(KernelMachine)已成为机器学习领域的热点研究问题。针对只具有离散属性的分类问题,在对合取范式进行深入分析的基础上提出了一族新的布尔核函数。利用这些布尔核函数,可以在布尔逻辑学习、决策树/决策规则学习以及基于项集的学习中,引入核机器技术。实验结果指出,使用结构简单而符合训练数据集特征的布尔核函数,有助于显著提高分类器的性能。
Currently there is a strong research interests for kernel machines, various kinds of kernels have been proposed by researchers for discrete data, such as set kernel, graph kernel, string kernel, tree kernel, and so on. However, little research has been done for Boolean kernel, which is defined on Boolean data. By taking utilization of Boolean kernels, it is possible to learn Boolean logic, decision tree, decision rule, and learning item sets with the help of kernel machines in the future. In this paper, by using some lemmas on kernel constructing, we present a family of Boolean kernels, and show that the current available Boolean kernels are all some certain instances in our family of Boolean kernels. We argue that if the structure of Boolean kernel represents the characteristic of training dataset, then the SVM with this kernel will have good classification performance on the testing dataset, and this is observed in our experiments.
出处
《微电子学与计算机》
CSCD
北大核心
2006年第8期213-215,共3页
Microelectronics & Computer
关键词
布尔核函数
核机器
分类器
Boolean kernel, Kernel machine, Classification
