Support Vector Machines

Question 1

What decides the direction of the separating hyperplane?

Answer 1

The weights

Question 2

What decides the distance from the origin of the separating hyperplane?

Answer 2

The bias

Question 3

Given a hyperplane, wX + b = 0,

how can we construct a parallel hyperplane?

Answer 3

A parallel hyperplane would have the form wX +b = c

Question 4

If I have two parallel hyperplanes,

wX + b = 0

wZ + b = 1

what is the distance between them?

Answer 4

1/L2_norm(w)

Question 5

What is the aim of an SVM?

Answer 5

To find a hyperplane that splits the two classes with the widest margin

Question 6

What are we trying to minimise using a Hard-margin SVM?

Answer 6

The inverse of the margin

We want the largest margin possible, so we try to maximise its inverse

Question 7

What are Support Vectors in a Hard-margin SVM?

Answer 7

Points that lie directly on the plane parallel to the separating boundary

Question 8

What is one way to find the minimum for SVMs?

Answer 8

By creating a dual problem and using quadratic programming.

Question 9

What are support vectors in a Soft-margin SVM?

Answer 9

Points that lie on the parallel hyperplane, in the margin, or are on the wrong side of the hyperplane

Question 10

How do we classify non-linear data using an SVM?

Answer 10

Modify the Quadratic Programming solution to use basis functions instead of the inner product