Decreasing training time for SVM



As far as I understand, in SVM, there’s a kernel trick which makes a new feature space from the available features (eg, x^2 + y^2 is created from x and y). But as the feature space increase the algorithm will require more time to do its processing. So what can we do to decrease train time for SVM?

Also, is it used often on a normal basis (real life scenarios)?


I am having the same problem, so I am looking at doParallel package in R, See if you can find something.