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 types of sequential minimal optimization(SMO) algorithms applied in solving Laplace-SVR with nonpositive kernels are proposed. The first algorithm is only designed for Laplace-SVR, and the second one regarding Laplace-SVR as a special case is done for a general purpose. Because of the difficulty of solving SVR with non-positive kernels, the presented algorithms have a certain theoretical and practical significance.