基于简单特征值问题的修正GEPSVM

Modified GEPSVM Based on Simple Eigen-values Problems

针对GEPSVM可能存在奇异性问题,提出一个改进的GEPSVM算法——SMGEPSVM。基于简单特征值问题,得到2个非平行过原点的超平面。与GEPSVM相比,SMGEPSVM不仅可以求解XOR问题,还可求解简单特征值问题,避免GEPSVM奇异性问题,测试精度好于GEPSVM,学习速度更快,其计算结果在人工数据集与UCI上得到了验证。

Aiming at the flaw of singular problems appearing in GEPSVM, this paper proposes an improved GEPSVM--SMGEPSVM. Two non-parallel surfaces are obtained by solving simple Eigen-values problems. Compared with GEPSVM, the approach can solve XOR problems, and it further solves the simple Eigen-value problems, avoids the singular problems appearing in GEPSVM, it has better test set correctness and faster training time. Computational results on artificial datasets and UCI indicate the better performance of SMGEPSVM.