Ali Ghodsi Lec 10, SVM, Kernel SVM

[@huwenhan]

Kernel SVM start @38:53 Soft Margin SVM start @58:01

[@MrWkuling]

Thx for uploading the videos. Will lectures 9 and 6 also be made available?

[@RahulPatel-gy3co]

Are all quadratic programming problems necessarily convex? As far as I know, the general QP (P) as min 1/2 x^TQx + c^Tx s.t. linear constraints is convex iff the Q (Hessian) is positive semi-definite.

[@asiphemzaza7471]

No, QP problems aren't necessarily convex. It is correct that Q must be positive semidefinite(PSD). This is the case here; the matrix S (at time 36:10) is certainly PSD by construction. Hence this is a convex quadratic program, with a solution guaranteed if the KKT theorem holds.