SVM:
1) in the case that data can be seperable
2) un-seperabel data need l1 regulazation
solving SVM with
Lagranga Dual function for optimation
Kernel for feature transformation
SMO for optimation of mutivariable function with constrain (the dual function)