The key problem of training support vector machines is how to solve quadratic programming problem, but for large training examples, the problem is too difficult.