SVM

Question 1

what constitutes a support vector in regular SVM

Answer 1

any point or vector that is on the margin, where either a . y . K(x,z) = 1 or -1 . Or yw^Tx - b = +-1

Question 2

What constitutes a support vector on a soft margin SVM

Answer 2

Any vector which is either:
-on the margin
-within the margin
-on the wrong side of the decision boundary and outside the margin

Question 3

what is the support vector algorithm (dual representation soft margin

Answer 3

-initalise a: // choose any value which satisfies constraints, any a^(n) where it equals 0
-Repeat for maximum iterations:
–select a pair a^(i) and a^(j) to update
–optimise loss function w.r.t. a^(i) and a^(j) while keeping all other pairs constant

Question 4

during the SVM soft margin algorithm, how do we optimise the loss function w.r.t the two selected LaGrange multipliers

Answer 4

when we select a^(i) . y^(i) and a^(j) . y^(j), all other La Granges are held constant. So we can have the equation in the image

The rest is apparently outside the scope of the module

Question 4

during the SVM soft margin algorithm, how do we optimise the loss function w.r.t the two selected LaGrange multipliers

Answer 4

when we select a^(i) . y^(i) and a^(j) . y^(j), all other La Granges are held constant. So we can have the equation in the image

The rest is apparently outside the scope of the module