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)

Read more »

When we try to use MLE to maximize marginal distribution of x. which is difficult to solve & x in a joint distribution with r (from graphic model: summing joint dist(x, y) get marginal dist).
We trun to optimaze the joing distribution. and we try to marginalize/kick out r (consider as latent variable) out and make our life easier
to solve the problem we need to use EM as solver.
my implementaion of EM algo:

Read more »