MA8701 General statistical methods

This was discussed with the reference group in a meeting on April 26.

The oral exam will take place on Monday May 13 and Tuesday May 14, 2019. The candidates have been given timeslots by email.

The oral exam needs to be passed with a 70% grade limit. Only pass/fail is possible.

The oral exam will last around 30 minutes (maximum 45 minutes), and

• will start with the candidate given a 10 minutes presentation (at the blackboard, student can bring notes) on one of the questions given below. The student can choose not to prepare a presentation, but then the lecturer will choose one of the five questions below and ask this as the first part of the exam.
• For the rest of the exam the student will answer questions (on the blackboard) from the lecturer.

An external sensor is present at the exam.

(What is Part 1-5? See Reading list in menu at the left.)

A: What is regularization? Choose one method from each of Part 1, 2 and 4 and explain (mathematics and practice).

B: What is a hyperparameter? What are important ingredients in choosing the value of a set of hyperparameters? Relate to key concepts from Parts 1, 2, 3 and 4.

C: What is a loss function? Many loss functions are motivated from the likelihood, elaborate. Relate the concept of loss function to one example from each of Parts 1 and 4.

D Design of experiments (DOE) can help to improve the learning obtained from your data in various ways. Give an overview over some situations where DOE is useful, explain the relevant methods and the corresponding cross-validation if necessary.

E: There are different kind of splines that are used in analysis of regression data. Give an overview of the topic as presented in Chapter 5 of Elements of Statistical Learning. What are regression splines and what are smoothing splines, and what is the difference?