Causality

Causality focuses on the study of causal relationships among the variables of a system of interest. Whereas standard machine learning typically learns correlations between features, causality aims at disentangling causes and effects. Understanding relations of cause and effect provides a better understanding of the system under study, offer improved explainability to the final user, and guarantees a better control over the system.

We study causality relying on the language of structural causal models (SCM), a formalism that relies on graphical models to encode variables of interest as nodes and causal relations as edges. Research in causality spans many subfields, such as causal discovery (how to learn causal model from data), causal inference (how to draw sound causal conclusions), causal transportability (how to perform causal inferences across models), causal explanations (how to exploit causality to provide explanations), causal decision-making (how to make optimal decision given a causal model). We specifically focus on causal abstraction, the problem of learnin and expoiting relationships between SCMs representing the same systems at different levels of resolution.