Sample Selection Based on K-L Divergence for Effectively Training SVM
Junhai Zhai, Chang Li · 2013
The computational time and space complexity of support vector machine (SVM) are O(n3) and O(n2) respectively, where n is the number of training samples. It is inefficient or impracticable to train an SVM on relatively large datasets. Actually, the removal of training samples that are not support vector (SVs) has no effect on constructing the optimal hyper plane. Based on this idea, this paper proposed a sample selection method which can efficiently choose the candidate SVs from original datasets. The selected samples are used to train SVM. The experimental results show that the proposed method is effective and efficient, it can efficiently reduce the computational complexity both of time and space especially on relatively large datasets.