I)raining support vector machines based on genetic algorithms

To overcome the vast computation of training SVM, fast training techniques are very useful and valuable. In this paper, we introduce an Improved GetBorder method which selects potential support vectors as reduced vectors. Then training SVM is equivalent to a quadratic programming of reduced vectors, whose size is very smaller comparing with the training data. Finally we apply genetic algorithms to optimize the quadratic programming of reduced vectors. Simulations illustrate that for linear separable training data, this novel method can train SVM effectively and efficiently. This method is also feasible even if the training samples scale is very large.