Week 3

Question 1

What is the definition of emperical risk?

Answer 1

Question 2

What is the definition of a unbiased risk estimator?

Answer 2

Question 3

What is a naive approach to emperical risk minimization?

Answer 3

Question 4

What is the definition of a model and a Emperical Risk Minimizer?

Answer 4

Question 5

What can be done to avoid the naive approach?

Answer 5

Question 6

Reframe a linear regression as an ERM.

Answer 6

Question 7

What is the definition of a partition?

Answer 7

Question 8

What is the definition of a step-function over a partition?

Answer 8

Question 9

What do you need to proof when finding the ERM of a step-function for regression.

Answer 9

Question 10

Proof the ERM of a step-function of a regression.

Answer 10

Question 11

What do needs to proved when proving the ERM of a step-function for classification?

Answer 11

Question 12

Proof the ERM of a step-function of a classification.

Answer 12

Question 13

What is the definition of a learning rule?

Answer 13

Question 14

What is the definition of a ERM learning rule?

Answer 14

Question 15

What should a good learning rule have?

Answer 15

Question 16

How are classification learning rules and regression learning rules linked?

Answer 16

Question 17

Why is a 0-1 cost function a bad idea?

Answer 17

Question 18

What is the definition of a pseudo-classifier?
What is the definition of a plug-in classifier

Answer 18

Question 19

What is the definition of a Empirical Convexified Risk Minimizer?

Answer 19

Question 20

What is the definition of Plug-in classification rule associated to ECRM.

Answer 20

Question 21

What is the idea of convex surrogate loss?

Answer 21

Question 22

Proof that a good regression is also a good classifier.

Answer 22