What is the difference between subgradient svm and kernel svm?
From my understanding subgradient svm is a linear classifier that uses hinge loss and kernel svm uses some kernel function for non linear classification. I was wondering, if subgradient is only a linear classifier, could I use a tree of non linear svms to separate non linear data? I would essentially do binary classification separate 1 class vs the rest then the child node of the tree would work on the rest and separate the next class from the rest and so on. Any general idea or feedback would be great.