论文部分内容阅读
提出2种用于求解非正定核Laplace-SVR的序列最小最优化(SMO)算法.第1种算法仅针对Laplace-SVR而设计;第2种算法将Laplace-SVR作为所要解决问题的一种特殊情况,使算法更具通用性.所提出的算法在保证收敛的前提下,使非正定Laplace-SVR能够达到比较理想的回归精度,具有一定的理论意义和实用价值.
Two kinds of sequence minimum optimization (SMO) algorithms for solving non-positive definite kernel Laplace-SVR are proposed. The first algorithm is designed only for Laplace-SVR. The second one takes Laplace-SVR as a special problem to be solved Which makes the algorithm more general. The proposed algorithm can make the non-positive definite Laplace-SVR achieve the ideal regression precision under the premise of ensuring convergence, which has a certain theoretical significance and practical value.