Prospectus

25-26,24-25

Level

MSc, 6 EC

Part of

MSc Computer Science

Lecturers

Saber Salehkaleybar

Introduction

In the field of machine learning, you may need to predict the impact of an action on an outcome by analyzing a huge amount of collected data. Unfortunately, this data is often of low quality, with missing records, unobserved confounders, and selection biases. As a result, answering questions about the impact of an action becomes more difficult than ever. This course covers mathematical tools to help you perform causal inference in big data. We will use structural causal models, and potential outcomes to formalize what causal effects mean, explain how to express these effects as functions of observed data, and use machine learning techniques to estimate them. We will also cover causal structure learning algorithms to recover causal relationships from data, and discuss applications of causality in other fields in machine learning (such as reinforcement learning and natural language processing).

Contents

In the course, we cover the following topics:

• Structural causal models (SCM), Potential outcome (PO) framework
• Interventions, Counterfactuals
• The tasks of causal discovery and causal inference (the bivariate case)
• Causal discovery (the multivariate case):
• d-separation, Markov equivalence class
• Constraint-based methods (IC, PC, and FCI)
• Score-based methods (GES)
• Learning from time-series (Granger causality, Directed information)
• Causal inference:
• Instrumental variables
• Natural experiments
• Back-door criterion, Front-door criterion
• Do-calculus
• IP weighting, Propensity scores
• Applications of causality