Hide menu

Causal Inference with Graphical Models

2019VT

Status Cancelled
School IDA-gemensam (IDA)
Division STIMA
Owner José M Peña

Will be given fall 2019 instead

  Log in  




Course plan

Goal

Causal inference comprises the study of cause and effect relationships, e.g. the conditions under which they can be elucidated from observations and/or be used to compute causal effects without actually performing interventions. More specifically, when can we determine from observations if our habits are the cause of a certain disease ? Or, when can we compute from observations the effect on our health of a prescribed treatment ? The goal of this course is to show the students how to answer these questions with the help of graphical models.

Since predicting the consequences of decisions or actions is necessary in many disciplines, it is not surprising that research on causal inference has a long tradition. Specifically, causal inference can be traced back to the work by Wright (1921), where path analysis was introduced for the first time. In path analysis causal relationships are represented with directed edges, and correlations due to unobserved common causes are represented with bidirected edges. Wright showed how to use such a graph-based model (a.k.a graphical model) to perform causal inference. Since then, the field has grown and matured: New graphical models have been proposed, new algorithms for causal effect identification have been developed, and algorithms for learning causal relationships from observations have been devised. Most of these results are reported in the books by Pearl (2009) and Peters et al. (2017). The goal of this course is to introduce the students to these works.

Recommended for

Students in the field of machine learning, artificial intelligence, or statistics.

Prerequisites

Basic statistics and probability theory.

Contents

- Path analysis (Wright's work).

- Graphical models (Pearl's work on Bayesian networks, and acyclic directed mixed graphs).

- Intervention calculus (Pearl's front-door criterion, back-door criterion, and do-calculus).

- Learning from observations (Peters' work on additive noise causal models).

Literature

Koller, D. and Friedman, N. Probabilistic Graphical Models: Principles and Techniques. MIT Press, 2009.

Pearl, J. Causality: Models, Reasoning, and Inference. Cambridge University Press, 2009.

Pearl, J., Glymour, M. and Jewell, N. P. Causal Inference in Statistics. A Primer. J. Wiley & Sons, 2016.

Peters, J. and Janzing, D. and Schölkopf, B. Elements of Causal Inference: Foundations and Learning Algorithms. MIT Press, 2017.

Organization

Lectures, seminars and computer labs.

Examination

Written report.

Lecturer

Jose M. Peña

Examiner

Jose M. Peña

Credits

6 HEC


Page responsible: Director of Graduate Studies