DAT465 / DIT654 Causality and causal inference

Jump to today

Welcome to Causality and Causal Inference!

Lp1 HT22 (7.5 hp)

Examiner & lecturer:

Fredrik Johansson, fredrik.johansson@chalmers.se
Computer Science and Engineering

Teaching assistant:

Lena Stempfle, stempfle@chalmers.se 
Computer Science and Engineering

Student representatives: 

  • Hanna Tärnåsen
  • Sanna Persson
  • Ruggero Barbarossa
  • Nicolas Audinet de Pieuchon

Announcements

  • Nov 14. Grades will be posted on Wednesday Nov 16 at the latest.
  • Oct 26. I just realized that we don't have ES52 booked today because it is exam week. 
    I'm available for questions regarding projects in my office EDIT 5482 and on Zoom https://chalmers.zoom.us/j/64381573608  (Passcode: 615736) 
  • Oct 21. On Wednesday the 26th next week, we will have another project workshop where you discuss your projects with your peers. 
  • Oct 12. On Wednesday the 19th next week, we will have a project (+ paper review) workshop where you discuss your projects/papers with your peers. There will be no lectures in the week of October 24–28.
  • Oct 6. If you want to do a MSc thesis with a supervisor from the DSAI division, you must request supervision by November 10. Information here 
  • Oct 6. Assignment 3 has now been posted.
  • Sep 30. Tomorrow's consultation will be available on Zoom as well as in person. We'll see how it works:  https://chalmers.zoom.us/j/69838307104, password: DAT465
  • Sep 20. For this week's topic on Structural causal models, the second part of last year's first module—Probabilistic graphical models— may be good for context. Those videos are posted in the module page.
  • Sep 12. A typo has been corrected in Assignment 1. The comment after Problem 1 about showing derivations should refer to part "B", not "C". The assignment has been updated. 
  • Sep 7. Clarification regarding the intro lecture on Potential outcomes & Experiments: At the end, I asked if a certain scenario was an example of Simpson's paradox. I was too unclear in my answer. No, it is not an example of Simpson's paradox. Because we are comparing the interventional distributions directly, the average effect can't be positive at the same time as it is negative in all subgroups. I will clarify in the module page as well. 
  • Sep 5. A clarification regarding lecture recordings has been added. Interactive elements (discussions/problem solving exercises on Follow-up meetings) will not be recorded. 
  • Sep 2. Assignment 1 has now been posted.
  • Aug 29. Slots for consultation hours have been added (see above)
  • Aug 10. The Canvas page has been created and the syllabus is being updated.

Schedule

Lectures: We meet in ES52 on Mondays (10:00-12:00am) Wednesdays (1:15–3:00pm).  Schedule: See TimeEdit

Recordings: Module Introduction lectures and non-interactive components of Follow-up sessions will be recorded and posted on the Canvas pages for each module. 

Consultation hours: Fridays  8:30–10:00am in EDIT 5482 and https://chalmers.zoom.us/j/69838307104, password: DAT465

Course purpose & structure

The aim for this course is to give knowledge and understanding of causality and causal inference from a mathematical, statistical and computational perspective. It should also provide the tools necessary for solving causal inference problems in practical applications. Students who complete this course should be able to recognize, define, and solve problems related to causality. This will involve becoming familiar with and using several paradigms (languages) of causality as well as methods for causal estimation and inference.

The content is structured into 8 modules, one for each week of the course. We begin by briefly covering necessary prerequisites, such as probabilistic graphical models, and by introducing structural definitions of causality. This will enable us to study sufficient conditions for inferring causal relationships between random variables. In the later half of the course, we study estimation of causal effects and policy evaluation. These are important topics in, e.g., epidemiology. At the end of the course, each student will present a research paper on a topic of their choice. See the overview below.

 

Schedule-crop-1.png

 

Sessions

  • Mondays are follow-up lectures
  • Wednesdays are introduction lectures

Modules

 The full schedule can be found in TimeEdit

Changes since 2021:

  • The introduction now places larger focus on machine learning and its relation to causality
  • Module have been rearranged for better flow through the course. Keep this in mind if watching old lectures

Course literature

The course is based on public material such as research papers and e-books available online. It will not follow a textbook week by week, but each module will be accompanied by slides, lectures and reading suggestions. If you would like to read a textbook on the subject, consider one of the following:

  • Morgan, Stephen L., and Christopher Winship. Counterfactuals and causal inference. Cambridge University Press, 2015. 2nd Edition.
  • Pearl, Judea. Causality. Cambridge university press, 2009. Available online through the Chalmers library
  • Peters, Janzing & Schölkopf, Elements of Causal Inference, MIT Press, 2017. Available online through the Chalmers library
  • Hernán MA, Robins JM (2020). Causal Inference: What If. Boca Raton: Chapman & Hall/CRC. 

Additional resources: 

Learning objectives and syllabus

On successful completion of the course the student will be able to:

Knowledge and understanding

  • Provide an overview over different frameworks for causality and causal inference
  • Describe important causal problems and give examples of their applications
  • Explain how problems of causal inference differs from other types of (statistical) inference problems
  • Account for common approaches to causal inference and the conditions under which they are accurate

Skills and abilities

  • Identify different types of causal inference problems in real-world applications
  • Analyze causal inference problems and estimate the plausibility of solving them under certain conditions
  • Implement and apply methods for causal inference appropriate for specific problems

Judgement and approach

  • Discuss pros and cons of different frameworks for causality and causal inference
  • Reflect over the fundamental limitations to the possibility of causal identification
  • Critically analyse and discuss research and applications of causal inference, in particular with respect to adjustment for confounding factors

Link to the syllabus: Study plan (Chalmers), Studieplan (GU)

Examination

The examination in this course will include five components comprising a) three written hand-in assignments, b) a smaller implementation project, c) a presentation of a research papers. These will take place throughout the entire course and are aimed to cover the full range of learning objectives. There will be no written final exam. Each problem-solving assignment will be awarded a score in the range 0–20, the paper review a score in the range 0–10, and the project a score in the range 0–30. For problem-solving assignments, it is not necessary to attempt "advanced problems" to achieve a passing grade or a "4" (0–15p). A grade corresponding to a "5" (16-20p) cannot be achieved without attempting the advanced problem.

Assignments are to be solved individually, as is the paper review. You are encouraged to discuss the material but to hand in your own solution. We expect a high degree of academic honesty in courses at this level. The project can be solved individually or in pairs, it's your choice. If you submit a solution before the deadline that does not meet the requirements for a passing grade, you will get some feedback and be asked once to correct the most important errors within a stipulated period of time. If you miss the deadline (or the resubmission deadline), the assignment solution will get the grade of Fail (U). It will be possible to resubmit failed assignments only in the last week of the course.

All instances of plagiarism, including but not limited to copying significant portions of text or code from the web will be reported and result in a failed assignment. 

For a passing final grade, each assignment should be completed with a minimum score of 40% and a total score, across all assignments, of 40%  (see grading thresholds below). To achieve a passing score, solutions are expected to have at most a few technical errors. To get a higher score, solutions/presentations should be correct, well explained, and more insightful.

Grading

  • For Chalmers and GU students (DAT465 / DIT654), a numerical scale is used (U, 3, 4, 5).

  • For PhD students, the grades for the course are  Pass (G)  or Fail (U)

The final grade will be based on the total sum of scores for all the assignments, using the following thresholds:

  • At least 80%  of the maximal: 5
  • At least 60–79%  of the maximal: 4
  • At least 40–59% of the maximal: 3

Course summary:

Date Details Due