论文部分内容阅读
核化一类硬划分SVDD、一/二类L2-SVM、L2支持向量回归和Ranking SVM均已被证明是中心约束最小包含球.这里将多视角学习引入核化L2-SVM,提出核化两类多视角L2-SVM(Multi-view L2-SVM),并证明该核化两类Multi-view L2-SVM亦为中心约束最小包含球,进而提出一种多视角核心向量机Mv CVM.所提出的Multi-view L2-SVM和Mv CVM既考虑了视角之间的差异性,又考虑了视角之间的关联性,使得分类器在各个视角上的学习结果趋于一致.人造多视角数据集和真实多视角数据集的实验均表明了Multi-view L2-SVM和Mv CVM方法的有效性.
One class of hard-partitioned SVDD, one / two L2-SVM, L2 support vector regression and Ranking SVM have all been proved to be the least constrained center-bounding sphere.In this paper, multi-view learning is introduced into corpus-based L2-SVM, Class multi-view L2-SVM, and prove that the two kinds of multi-view L2-SVMs are also center-constrained least-contained balls, and then a multi-view core vector machine Mv CVM is proposed. Multi-view L2-SVM and Mv CVM not only consider the difference between perspectives, but also consider the correlation between perspectives, which makes the learning results of the classifier consistent in all perspectives.The artificial multi-view data sets and Experiments with real multi-view datasets demonstrate the effectiveness of Multi-view L2-SVM and Mv CVM methods.