计算机应用与软件2016，Vol.33Issue(5)：239-241,264,4.DOI:10.3969/j.issn.1000-386x.2016.05.060
相关向量机基函数和超参的协同优化
COLLABORATIVE OPTIMISATION OF BASE FUNCTION OF RELEVANCE VECTOR MACHINE AND SUPER PARAMETERS
摘要
Abstract
Traditional relevance vector machine has the conflict among training error,sparseness of weight matrix and zero-approaching of log marginal likelihood function.To solve this problem,in this paper we present to utilise receiver operation curve to carry out collaborative optimisation on parameters of relevance vector machine and kernel function.According to the accuracy rate of model classification we determine proper kernel function.By introducing the classification accuracy rate of model at 5 percent false positive rate we improve the marginal likelihood function of super parameters.In order to ensure the maximisation of weight matrix sparseness,we choose the optimal relevance vectors combination through the threshold of marginal likelihood function.The cross-validation algorithm and the receiver operation curves of all cross models are used to estimate the optimal super parameters of relevance vector machine.Moreover,we use vehicle yaw angular velocity to test the optimised model,results show that the training time of the proposed algorithm is a little bit longer,but the test time is obviously shorter than traditional estimation algorithm,and the classification performance of the optimised model is improved dramatically.关键词
相关向量机/基函数/超参/协同优化/ROC 曲线Key words
Relevance vector machine/Kernel function/Super parameters/Collaborative optimisation/Receiver operation curve分类
信息技术与安全科学引用本文复制引用
张名芳,付锐,郭应时,程文冬..相关向量机基函数和超参的协同优化[J].计算机应用与软件,2016,33(5):239-241,264,4.基金项目
国家自然科学基金项目（61374196，51178053）；教育部长江学者和创新团队发展计划项目（IRT1286）。 （）
