The Swiss engineering master's degree

Automatising causal inference is one of the main challenges for making artificial intelligence (AI) reliable and thus really useful in the real world, as more and more emphasised by scientists and practitioners:

“Machines’ lack of understanding of causal relations is perhaps the biggest roadblock to giving them human-level intelligence.” (Judea Pearl, Turing Award winner and AI pioneer.)

“Causality is very important for the next steps of progress of machine learning.” (Yoshua Bengio, Turing Award winner and “Godfather of Deep Learning”.)

“Causal AI is a key enabler of the next wave of AI, where AI moves toward greater decision automation, autonomy, robustness and common sense.” (Gartner, Analyst Firm.)

The list of applications that can be addressed by causal AI is long and important, e.g.: (medical) treatments; marketing strategies; disparity/fairness/discrimination and AI ethics more in general; information fusion; explainability; robustness; various applications in economics, medicine, epidemiology, the social sciences, etcetera.

In order to having access to these capabilities, the module will introduce students with the most important concepts in causal inference. In particular, after a review of concepts in probability and graph theory, it will focus on the treatment of interventions, counterfactual, and mediation analysis. Lectures will be constantly accompanied by examples and made very concrete through exercises based also on software for causal inference. 

Prerequisites

Basics of probability theory and machine learning.

Learning Objectives

This module will enable students to get a solid understanding of the most important concepts and algorithms in causal inference, and to have hands-on experience on the practical use of causal inference. At the end of the module, students will be able to model problems in a causal fashion and have them solved by state-of-the-art algorithms. They will be able to address many types of applications that are not accessible by engineers with a machine learning curriculum alone and that are more and more relevant in the industry.

Contents of Module

The module will cover the following topics. Introduction: causal inference vs machine learning; review of elementary concepts in probability and statistics; Bayesian networks. Interventions: observational vs randomised controlled studies; causal effects; causal inference in linear systems. Counterfactuals: structural causal models; personal decision making; discrimination; attribution; mediation.

The topics above will be constantly backed up with practical examples and use of software to make inference with structural causal models. Students will eventually be required to work on a (simulated) applied project where they will test their new competences all the way through the modelling of a problem to its solution and evaluation.

Teaching and Learning Methods

• Lectures / presence
• Tutorial / presence
• Self-study

Literature

Judea Pearl, Madelyn Glymour, Nicholas P. Jewell. Causal Inference in Statistics, a Primer. Wiley, 2016.