SVM

Question 1

What is a margin? What are the support vectors? Which is the minimun number of support vectors required? What is a Hard-SVM?

Answer 1

7 / 4-7

Question 2

Describe generally the equivalent formulation for Hard-SVM, the one with homogeneous halfspaces

Answer 2

7 / 9

Question 3

When Soft-SVM are used? Describe them and also write down the optimization problem they solve

Answer 3

7 / 10-12

Question 4

Write down the equivalent formulation for Soft-SVM (hinge loss). How can be solved? Try to describe in detail the algorithm.

Answer 4

7 / 12-15

Question 5

In Hard-SVM there exits a dual formulation. For this case, what is the only thing that needs to be computed?

Answer 5

7 / 16

Question 6

What is a kernel function? Describe the general procedure on how to apply the transformation on the training set S and then find the correct prediction.

Answer 6

7 / 18-19

Question 7

What is the kernel trick? What are the most
common kernels?

Answer 7

7 / 20-23

Question 8

Define the Degree-Q polynomial kernel

Answer 8

7 / 24

Question 9

Define the sigmoid kernel

Answer 9

7 / 23

Question 10

Define the Gaussian-RBF kernel

Answer 10

7 / 25

Question 11

How do we choose the kernel? What the Mercer’s condition says?

Answer 11

7 / 26

Question 12

How to use SVM for regression? What is the function to minimize? What is the final model produced? What is the new definition of support vectors?

Answer 12

7 / 27-28