A brief introduction to causal discovery

Abstract: As a fundamental task, determining causality (also referred to as causal discovery) is needed in multiple disciplines of science. For many real-world applications, causality plays an important role in predicting the effects of a structural change. To obtain causal relations, it is required to properly design some specific interventions and randomized experiments. But in many cases, such experiments are quite challenging or even impossible due to the limitation of ethical issues, funding, or technical feasibility. Therefore, discovering causality purely based on observational data has attracted more attention recently.

This talk will make a brief introduction to the existing causal discovery approaches, from traditional methods, such as constraint-based and score-based, to more recent methods using functional causal models as well as the ones leveraging machine learning algorithms. This talk will also discuss some practical issues in causal discovery applications and challenges in evaluating causal discovery methods.

Bio: Ruibo Tu is a doctoral student in the Division of Robotics, Perception, and Learning at KTH Royal Institute of Technology. He is supervised by Prof. Hedvig Kjellström at KTH and co-supervised by Dr Cheng Zhang at Microsoft Research Cambridge, and Prof. Kun Zhang at Carnegie Mellon University. His research interests include causal discovery, fairness, and missing data problems.

Link to the profile of Ruibo Tu