Fast Prediction for Large-Scale Kernel Machines

Neural Information Processing Systems 

Kernel machines such as kernel SVM and kernel ridge regression usually construct high quality models; however, their use in real-world applications remains limited due to the high prediction cost.