中心向量夹角间隔正则化核向量机

Regularized Core Vector Machine with Central Vector-Angular Margin

针对大数据集如何有效地进行训练的问题,基于最大向量夹角间隔分类器(maximum vector-angular margin classifier,MAMC),提出了求解最优向量d的不同方法来得到中心向量夹角间隔分类器(central vector-angular margin classifier,CAMC),进而证明了CAMC等价于最小包围球问题(minimum enclosed ball,MEB).但是鉴于MEB对参数的敏感性,又提出了正则化核向量机(regularized core vector machine,RCVM),将CAMC与RCVM结合得到中心向量夹角间隔正则化核向量机(regularized core vector machine with central vector-angular margin,CAMCVM).基于基准数据集的实验表明,CAMC具有更好的分类性能且CAMCVM可以有效快速地训练大规模数据集.

For effective training on large datasets,we propose an alternate method to find the optimal vector,d, using the central vector-angular margin classifier （CAMC）,which is based on the maximum vector-angular margin classifier.The CAMC can be considered to be equivalent to the corresponding minimum enclosing ball （MEB）problem.However,we have found that the MEB is very sensitive to the selection of the trade-off pa-rameter,so we propose using a regularized core vector machine （RCVM）.By connecting the CAMC to the RCVM,we obtain a central vector-angular margin regularized core vector machine （CAMCVM）.Experimen-tal results from the UCI datasets show that the CAMC has a better generalized performance,while the CAM-CVM can be used for effective training on large datasets.