>>14301253Its been awhile since I looked at the formulations for SVM, but in general optimizing convex functions over convex sets is efficient because local minima <--------> global minima. There are different algorithms for solving this, one example is you can convert a constrained optimization problem to an unconstrained one by penalizing constraint violation (see barrier method), then you can optimize the unconstrained problem using you're favorite flavor of gradient descent.
