A part-versus-part method for massively parallel training of support vector machines

Bao-Liang Lu, Kai-An Wang, M. Utiyama, H. Isahara
2004 IEEE International Joint Conference on Neural Networks (IEEE Cat. No.04CH37541)  
This paper presents a part-versus-part decomposition method for massively parallel training of multi-class support vector machines (SVMs). By using this method, a massive multiclass classification problem is decomposed into a number of twoclass subproblems as small as needed. An important advantage of the part-versus-part method over existing popular pairwkeclassification approach is that a large-scale two-class subproblem can be further divided into a number of relatively smaller and balanced
more » ... wo-class subproblems, and fast training of SVMs on massive multi-class classification problems can he easily implemented in a massively parallel way. To demonstrate the eITectiveness of the proposed method, we perform simnlations on a large-scale text categorization problem. The experimental results show that the pmposed method is faster than the existing pairwise-classification approach, better generalization performance can he achieved, and the method scaler up to massive, complex multi-class classification problems.
doi:10.1109/ijcnn.2004.1380009 fatcat:g3cns7lvzvf4hfszlh5lk7awhm