Lecturer: Michael Gutmann (contact)
Teaching assistant: Vaidotas Simkus (contact)

Recent announcements

Apr 8, 2022 The Course Enhancement Questionnaire is available on Learn -> Have your Say or via this link. The survey is open until 17th April 2022 at 23:59. You will need to be logged into your Office 365 account to access the survey. All responses are anonymous.
Mar 27, 2022 HMM repo updated with a demo/exercise on learning for HMMs.
Mar 14, 2022 Uploaded solutions to tutorial 7 and added a link to tutorial 8 (a jupyter notebook on VAEs).
Mar 14, 2022 Clarified and expanded the solution to Exercise 3c in Tutorial sheet 5.
Mar 11, 2022 Quiz 3 opens Friday March 18 at 3pm, closes Monday March 28 at 4pm Wednesday March 30 at 4pm.

Content

The course provides you with a thorough understanding of statistical/probabilistic machine learning methods that are used in a wide range of different fields with a focus on unsupervised learning. The course covers four main topics:

  1. probabilistic graphical models,
  2. exact inference,
  3. learning,
  4. approximate inference and learning.

The course material is available here and optional material here.

Assessment

Overview

  • The course is assessed by a written exam at the end of the course (75% of final mark) and three quizzes during the semester (25% of final mark). There is no assignment.
  • The quizzes check your engagement with the course and are typically easier than the exam.
  • The exam date is announced here and past exams are available here (search for probabilistic modelling and reasoning).
  • PMR follows the Common Marking Scheme of the university and the school-wide academic conduct policy.

Quizzes

  • The quizzes open 1 week prior to their due date. They are time-limited and you will you have 1 attempt for each quiz.
  • The maximal time to complete a quiz is 60 minutes. The 60 mins will allow you to consult the slides or tutorials, but you will still need to study the material beforehand. Otherwise, you will likely run out of time.
  • The quizzes are available on gradescope. Please access it via Learn.
  • The quizzes will be autosaved and you can revise them as often as you want but once you have started a quiz, you will have to finish it within 60 minutes after you started.
  • When a quiz “closes at 4pm”, it means that you can work on it until 4pm.

:heavy_check_mark:Test quiz: Opens Feb 06 at 3pm, closes Friday Feb 11 at 4pm Thursday Feb 17 at 6pm. This is to check that you can access gradescope and allows you to familiarise yourself with the types of questions asked in the quizzes. It takes less than five minutes to complete. Please get in touch with Michael in case of access problems or questions.

:heavy_check_mark:Quiz 1: Opens Friday Feb 11 at 3pm, closes Friday Feb 18 at 4pm (end of week 5). The quiz is on directed and undirected graphical models. It covers the material on the slides until (and including) “Undirected Graphical Models II”, and the exercises discussed in tutorials 1 and 2.

:heavy_check_mark:Quiz 2: Opens Friday March 4 at 3pm, closes Friday March 11 at 4pm (end of week 7). The quiz is on inference and message passing. It covers the material on the slides “Factor Graphs” and “Exact Inference”, and the exercises discussed in tutorial 4.

:heavy_check_mark:Quiz 3 Opens Friday March 18 at 3pm, closes Monday March 28 at 4pm (beginning of week 10) Wednesday March 30 at 4pm. The quiz is on learning. It covers the slides Basics of Model-Based Learning, Factor Analysis and Independent Component Analysis, Intractable Likelihood Functions, and Estimating Unnormalised Models by Score Matching, as well as Tutorial 6, Exercises 1-9.
Note: you will have to do some calculations by hand. So please have pen and paper ready.