A Study of Kernel Select for the Relevance Vector Machine
首发时间:2013-12-19
Abstract:In this paper a method of the kernel select for the relevance vector machine is proposed. Former researches of RVM usually focus on the advantages of the algorithm in sparsely for less opportunities of over learning, how to accelerate the algorithm in order for needs such as large scale sets, and how to improve the precision of the algorithm by the combination of other algorithms such as PCA or SVM. The parameters of the kernel functions are regarded more important than these functions themselves, so that how to select the appropriate kernel and its parameters is confused for the application of RVM. In this paper, a method for kernel select for the RVM is proposed. Firstly, by simulation experiment, the relationship between σ in Gaussian kernel function and the regression result is discovered. Secondly, by analysis the reason of the experimental result, a common principle for kernel select is presented. Finally, on the principle, a composite kernel and the relation between different parameters and the number of RVs and the error rate in regression are tested by simulation experiment.
keywords: Pattern Recognition Relevance Vector Machine Kernel Function
点击查看论文中文信息
相关向量机核函数选择研究
摘要:文章提出一种相关向量机核选择方法。现有针对相关向量机的研究主要集中在算法稀疏性以防止过学习、加快算法运算速度以适应大数据集等应用及通过与主成分分析、支持向量机等算法结合使用提高算法精度上,一般认为,较之核函数的参数选择,核函数形式本身并不重要。在实际应用中,对核函数及参数的选定并无依据。文章提出一种核选择方法。首先,通过仿真实验验证,找出在相关向量回归模型中,高斯核核参数σ与回归结果的关系;然后分析该实验结果,提出通用的相关向量机核选择原则。并以此原则为指导,给出混合核函数模型,分析其在不同参数下与相关向量个数及错误率关系,并通过仿真实验验证了该推断。
基金:
论文图表:
引用
No.****
同行评议
共计0人参与
勘误表
相关向量机核函数选择研究
评论
全部评论0/1000