中文摘要：

在综合考虑L1和L2多核判别分析的优点基础上,引入弹性正则化.以预定内核函数的线性组合为基础,结合混合范数正则化函数平衡核权重的稀疏性和非稀疏性,提出了一种基于半无限规划的弹性多核判别分析学习算法（EM-KDA）,该算法应用半无限规划算法求解弹性多核判别分析,并通过混合正则化来实现核的自学习.在不同数据集上的实验验证了算法的有效性,实验结果表明：该方法能够平衡L1和L2多核判别分析的稀疏性和非稀疏性,可以尽可能地利用基核的信息;与其他多核判别分析方法相比,具有更好的性能.

英文摘要：

On the basis of considering the advantages of L1 and L2 multi‐core discriminant analysis , elastic regularization was introduced .A kind of learning algorithm was presented based on the semi‐infinite programming of flexible multi‐core discriminant analysis （EM‐KDA ） ,the characteristics of the algorithm was based on a linear combination of the predefined kernel function ,utilizing mixed norm regularization function to balance the sparsity of kernel weights ,applying semi‐infinite program‐ming algorithm to solve the flexible multi‐core discriminant analysis ,through the mixed regularization to achieve kernel self‐learning .The experimental results in different data sets demonstrated the effec‐tiveness of the proposed algorithm and experiment results show that the method can balance the nature of sparsity and non‐sparsity between L1 and L2 multicore discriminant analysis ,and can use basis ker‐nel information as much as possible with a better performance compared with other multi‐core discrim‐inant analysis .