基于支持向量机的核学习研究.pdfVIP

摘 要 支持向量机作为一种发展成熟的机器学习算法, 已在各领域得到广泛应用, 但核参数的选择一直是影响支持向量机算法效果的核心问题. 因为计算时间复杂 度低且表现效果良好, 独立于支持向量机的核学习准则在近年来广受青睐. 本文介绍了两种典型的核学习准则, 即基于核对齐思想的中心化核排列(Cen- tered Kernel Target Alignment, CKTA) 和基于聚类思想的核余弦相似度 (Cosine Similarity of Kernel, CSK), 并分别对它们进行了改进. 针对CKTA 将正常样本点 和离群点同等对待, 可能会影响样本在特征空间中的可分性, 提出通过训练集对 应的近邻矩阵中同类样本的比例来重构理想核矩阵, 得到了新的局部中心化核 排列(New Local Centered Kernel Target Alignment, NLCKTA), 并基于最大化NL- CKTA 的取值推导出了其对应的多核学习方法. 针对CSK 取值不具有理论上的 非负性而不易推广到多核学习的问题, 通过将其同类平均余弦相似度拆分为正 类平均余弦相似度和负类平均余弦相似度得到了新的核余弦相似度(New Cosine Similarity of Kernel, NCSK), 并利用基本核函数对应的NCSK 值的相对大小给出 了其多核学习方法. 仿真实验结果表明, NLCKTA 在单核和多核学习能力上比起CKTA 有明显 提升,NCSK 保持了CSK 的单核学习能力, 在多核学习表现上整体也优于CKTA. 关键词: 支持向量机, 核函数, 核学习准则, 核参数选择, 多核学习 i ABSTRACT Asawell-developedmachinelearningalgorithm,SupportVectorMachine(SVM) hasbeenwidelyusedinvariousfields,buttheselectionofkernelparameterhasalways been the core problem affecting the effect of SVM algorithm. And the kernel learning criterion independent of SVM is extensively favored in recent years because of its low computationalcomplexity andgoodperformance. Inthispaper,weintroducetwotypicalkernellearningcriteria: theCenteredKer- nelTargetAlignment(CKTA)andtheCosineSimilarityofKernel(CSK),thenimprove them respectively. In view of the fact that CKTA treats normal sample points and out- liers equally, which may affect the separability of samples in the feature space. There- fore, the ideal kernel matrix is reconstructed by the proportion of samples belonging to the same class in the nearest neighbor matrix corresponding to the training set, and a new kernel learning method named New Local Centered Kernel Target Alignment (NLCKTA) is obtained. Based on maximizing the value of NLCKTA, the correspond- ing multiple kernel learning method is derived. To solve the problem that the value of CSK does not ha