基于Gram-Schmidt正交化和HSIC的核函数选择方法  

A Kernel Selection Method Based on Gram-SchmidtOrthogonalization and HSIC

作　　者：高雅田[1] 贾斯淇 GAO Ya-tian;JIA Si-qi(Dept.of Computer and Information Technology,Northeast Petroleum University,Daqing 163318,China)

机构地区：[1]东北石油大学计算机与信息技术学院,黑龙江大庆163318

出　　处：《计算机技术与发展》2024年第6期148-154,共7页Computer Technology and Development

基　　金：黑龙江省高等教育教学改革项目(SJGY20210149)。

摘　　要：核方法是一种解决非线性、异构数据的有效方法,核函数的选择问题是核方法中的一个重要课题,对于不同的应用问题,如何选择合适的核函数还没有足够的理论基础,不适当的核函数选取会降低核方法的性能。由此,提出了一种基于Gram-Schmidt正交化(GSO)和Hilbert-Schmidt独立准则的核选择方法(HSIC-GSO),该方法考虑了核函数选择过程中存在的不相关冗余信息。首先,利用GSO消除核函数之间的冗余信息;然后,使用HSIC度量核函数与理想核之间的相似性;最后,得到一组判别能力强、多样性大的基核函数。实验结果表明,HSIC-GSO方法选择的核函数泛化性好,并且提高了MKL的分类性能,验证了所提方法的有效性。Kernel methods are an effective method for solving nonlinear and heterogeneous data,and the selection of kernel functions is an important issue in kernel methods.For different application problems,there is not enough theoretical basis for selecting the appropriate kernel,and improper kernel selection can degrade the performance of kernel methods.A kernel selection method based on Gram-Schmidt orthogonalization(GSO)and Hilbert-Schmidt independence criterion(HSIC-GSO)is proposed,which considers the irrelevant redundant information present in the kernel function selection process.Firstly,GSO is used to eliminate redundant information between kernel functions.Then,HSIC is used to measure the similarity between the kernel function and the ideal kernel.Finally,a set of kernel with strong discriminative ability and high diversity is obtained.The experimental results show that the HSIC-GSO method has good kernel generalization and improves the classification performance of MKL,verifying the effectiveness of the proposed method.

关 键 词：多核学习 核函数选择 不相关冗余信息 Gram-Schmidt正交化 Hilbert-Schmidt独立准则 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]