摘要
为了给模式分类和维度约简提供有效的手段,在综合L1-MKDA和L2-MKDA两种多核KDA优点的基础上,提出一种以预定内核函数的线性组合,并结合混合范数正则化函数实现核权重的稀疏性和非稀疏性平衡的新型弹性多核判别分析方法(semi-infinite-programming-based flexible multi-kernel discriminant analysis,SFMKDA).该方法用半无限规划方法求解弹性多核判别分析算法,并通过混合正则化实现核的自学习.在不同数据集上的实验结果表明:S-FMKDA比目前常见的KDA、KDAP、KDAG、L1-MKDA、L2-MKDA、UMKDA核判别分析方法的精度提高5%.
In order to provide effective means for pattern classification and dimension reduction and stem from the advantages of two kinds of multi-kernel namely L1-MKDA and L2-MKDA,a new type of semi-infinite-programming-based flexible multi kernel discriminant analysis method was proposed,which is based on a linear combination of the predefined kernel function,and can utilize mixed norm regularization function to balance the sparsity of kernel weights. It applys semi-infinite programming algorithm to solve the elastic multi-core discriminant analysis,and achieves nuclear self-learning through the mixed regularization. Finally,the experimental results for different data sets demonstrate that the accuracy of the proposed algorithm is 5% better than those of KDA、KDAP、KDAG、L1-MKDA、L2-MKDA and UMKDA.
出处
《西南交通大学学报》
EI
CSCD
北大核心
2015年第6期1122-1129,共8页
Journal of Southwest Jiaotong University
基金
国家自然科学基金资助项目(61573171
51108209
61203244)
交通运输部信息化项目(2013-364-836-900)
江苏高校优势学科建设工程资助项目(PAPD)
全国统计科学研究项目(2014596)
江苏省自然科学基金资助项目(BK20140570)
江苏省六大人才高峰资助项目(DZXX-048)
关键词
多核
判别分析
范数
正则化
半无限
规划
稀疏性
multi-kernel
discriminant analysis
norm
regularization
semi-infinite
programming
sparsity