Incremental SVMs and Their Geometrical Analyses
Takemasa Yamasaki, Kazushi Ikeda · 2006
A support vector machines (SVM) is known to result in a quadratic programming (QP) problem, which requires a large computational complexity. Two incremental or iterative SVMs are proposed and analyzed from the geometrical viewpoint. One utilizes the fact that only effective examples are necessary and sufficient to obtain the SVM solution and update the effective set iteratively. This produces the same solution as the SVM in batch mode, however, it is not easy to implement. Hence, the other method stores the set of support vectors, instead. Both methods have the linear complexity in average and the learning curve reciprocal to the number of examples