PhD defense Samuel Reyd: From Observations to Explanations: Exploring Methodologies for Identifying Relevant Actual Causes in Complex Systems

Jury

• Hana Chockler, Professor, King’s College London (Reviewer)
• Sven Tomforde, Professor, Kiel University (Reviewer)
• Émilie Devijver, CNRS Researcher, Université Grenoble Alpes (Examiner)
• Marianne Clausel, Professor, Université de Lorraine (Examiner)
• Ada Diaconescu, Associate Professor, Télécom Paris (Thesis Director)
• Jean-Louis Dessalles, Professor Emeritus, Télécom Paris (Thesis Director)

Abstract

Complex systems such as multi-agent and cyber-physical systems often exhibit behaviors that are hard to understand, even for their designers. In this thesis, we study how causal models can be used to explain specific events in such systems, answering questions of the form “why did this happen?”. We approach this problem as a three-step pipeline: modeling the system causally, identifying the actual causes of an event, and filtering these causes to retain those most relevant to the user and the context.

The core of the presentation will focus on the identification step, in particular on approximate search algorithms with adjustable precision that recover the complete set of actual causes in general causal models, released as an open-source Python module. We will also discuss the choice of abstraction scale when modeling complex systems, through the lens of causal emergence.