摘要
核函数是核方法的重要组成部分,设计得好坏直接影响模型的分类效果,高斯核函数以其优良的特性被广泛应用,但高斯核参数的优化十分困难.针对此问题,使用核目标度量准则制定目标函数,将问题转化为求极小值的最优化问题,利用小批量梯度下降法求解目标函数.在十六组机器学习领域常用的数据集上进行测试,实验结果表明,该方法均具有最短的训练时间和较高的分类准确率.
Kernel function is an important component of kernel-based methods and the design quality directly exerts influences on the classification of the model.Gaussian kernel function has been widely used because of its excellent properties.However,research shows that the optimization of Gaussian Kernel Function is a quite challenging problem.In order to solve the problem,the kernel target alignment is used to formulate the objective function,then the problem is transformed into the optimization for minimum value,and finally mini-batch gradient descent method is employed to solve objective function.Generally speaking,by evaluating the empirical performance of the proposed method on sixteen diverse datasets commonly applied in the field of machine learning,the experimental results show that the proposed method is more effective with the shortest training time and higher classification accuracy.
作者
肖玉麟
XIAO Yulin(School of Big Data and Artificial Intelligence,Fujian Polytechnic Normal University,Fuqing,Fujian 350300,China)
出处
《福建技术师范学院学报》
2023年第2期149-155,共7页
JOURNAL OF FUJIAN POLYTECHNIC NORMAL UNIVERSITY
关键词
核方法
高斯核函数
核目标度量准则
小批量梯度下降法
kernel-based methods
Gaussian kernel function
kernel target alignment
mini-batch gradient descent method
