SVM Incremental learning are known to result in a quadratic programming problem, that requires a large computational consumption. To reduce it, this paper considers, from the geometrical point of view, hull vectors and center vectors. The given algorithm is based on utilizing the result of previous training effectively and retraining the most important samples(hull vectors) for incremental learning to reduce the computational cost. In the process of incremental learning,the hull vectors of the previous training and the newly added samples constitute the current trainging sample, the center vectors is used to remove noise sample from training sample and adjust the classification hyperplane farther. The experimental results indicate that the algorithm has better performance than other conventional SVM incremental algorithm when dealing with large training set.
[1]
Wu Tie-jun.
Fast incremental learning algorithm of linear support vector machine based on hull vectors
,
2006
.
[2]
Zhou Wei.
An Analysis of SVMs Generalization Performance
,
2001
.
[3]
Xiaoou Li,et al.
Support vector machine classification for large data sets via minimum enclosing ball clustering
,
2008,
Neurocomputing.
[4]
Luc Vincent,et al.
Morphological grayscale reconstruction in image analysis: applications and efficient algorithms
,
1993,
IEEE Trans. Image Process..
[5]
Yizhuo Zhang,et al.
Constructing Multiple Support Vector Machines Ensemble Based on Fuzzy Integral and Rough Reducts
,
2007,
2007 2nd IEEE Conference on Industrial Electronics and Applications.