Postgraduate Course: Methods for Causal Inference (INFR11207)
|School||School of Informatics
||College||College of Science and Engineering
|Credit level (Normal year taken)||SCQF Level 11 (Postgraduate)
||Availability||Available to all students
|Summary||Causal inference is an important emerging area in AI and data science allowing us to move away from merely associational statements towards cause-effect statements. Being able to develop and/or apply causal inference techniques has broad applications in social and biomedical sciences, e.g., in answering questions such as "How effective is a given treatment for curing/preventing a disease?" or "Which genetic variants can causally increase the risk of disease and hence be targeted by drugs?" or "What economic policy could have prevented the 2008 financial crisis?"
In this course we develop causal inference techniques to address the above questions. This is a relatively advanced course and students are expected to be familiar with foundations of probability, statistics and calculus (see "Other Requirements" box)
The aims and the structure of the course are as follows:
- Estimating causal effects: Why correlations alone are misleading?
- Randomised trials vs observational data
- Part I: Causal Effect Estimation
- Rubin's framework: Potential outcomes with observed and unobserved confounders
- Pearl's framework: Structural causal models with observed and unobserved confounders
- Computer simulations and numerical exercises in Python
- Part II: Causal Discovery
- Constraint-based algorithms and Score-based algorithms
- Functional Causal Models
- Computer simulations and numerical exercises in Python
Teaching of the theory is followed by illustrative examples from biomedicine and social sciences, together with appropriate computer simulations and numerical exercise.
Entry Requirements (not applicable to Visiting Students)
|| It is RECOMMENDED that students have passed
Machine Learning and Pattern Recognition (INFR11130) AND
||Co-requisites|| It is RECOMMENDED that students also take
Probabilistic Modelling and Reasoning (INFR11134)
||Other requirements|| This course is open to all MSc students at School of Informatics, including those on joint degrees. The course is also available to MSc students at School of Mathematics (e.g. MSc Statistics and Data Science).
Required prerequisite: Students should have passed MLPR (INFR11130) or Probability (MATH08066) or an equivalent course indicating facility with probabilistic models and mathematical thinking.
1. Linear algebra: Vectors, matrices: addition, multiplication, inversion, diagonalization (eigenvectors).
2. Special functions: properties and combination rules for logarithm and exponential.
3. Calculus: Differentiation and integration.
4. Probability theory: Discrete and continuous univariate random variables. Expectation, variance. Univariate Gaussian distribution. Joint and conditional distributions.
Students should be familiar with programming in a modern object-oriented language, ideally Python which is the course language.
Information for Visiting Students
|Pre-requisites||Same as "other requirements".
|High Demand Course?
Course Delivery Information
|Academic year 2021/22, Available to all students (SV1)
|Learning and Teaching activities (Further Info)
Lecture Hours 20,
Seminar/Tutorial Hours 5,
Summative Assessment Hours 2,
Revision Session Hours 2,
Programme Level Learning and Teaching Hours 2,
Directed Learning and Independent Learning Hours
|Additional Information (Learning and Teaching)
Feedback on asignments are provided during the lectures and tutorials
|Assessment (Further Info)
||- Questions on material during the lectures will be discussed
- Two feedback sessions within lecture hours, after each of the two course assignments (one formative, one summative)
- Feedback on the work during tutorials
- Weekly online quiz on Learn, with solutions provided after answering the questions (not assessed)
- Course forum on Learn for students to ask questions
||Hours & Minutes
|Main Exam Diet S2 (April/May)|| Methods for Causal Inference (INFR11207)||2:00|
On completion of this course, the student will be able to:
- Explain the difference between causal and associational estimation and justify why causal inference techniques are necessary to derive meaning from observational data
- Explain the difference between randomised trials vs observational studies related to public health and other types of data more generally
- Learn and apply foundational causal estimation techniques using two major frameworks: (i) Rubin's Potential Outcomes and (ii) Pearls Structural (graphical) causal models to simulated examples and real world data, in the presence of observed and unobserved variables
- Explain different types of causal discovery algorithm, learn their underlying assumptions and short-comings, and be able to apply them to data using available software.
- Be able to modify/repurpose a current technique in order to apply it to a particular problem of interest.
|Books that may be useful but not required:|
1) Causal Inference in Statistics: A Primer (Pearl, Glymour, Jewell, 2016).
2) Elements of Causal Inference: Foundations and Learning Algorithms (by Peters, Janzing and Schölkopf
3) More advanced: Causality (Pearl, 2009)
|Graduate Attributes and Skills
||problem-solving, critical/analytical thinking, independent learning, cross-cultural or cross-disciplinary communication
|Keywords||Causal inference,causal discovery,data science,probability and statistics,Machine Learning,MCI
|Course organiser||Dr Ava Khamseh
Tel: (0131 6)51 1426
|Course secretary||Ms Lindsay Seal
Tel: (0131 6)50 2701