Causal discovery and causal inference with background knowledge

Specialeforsvar: Tobias Strømgren

Titel: Causal discovery and causal inference with background knowledge

Abstract: Understanding causal relationships is crucial for distinguishing causation from mere correlation. This importance spans various fields, including medicine, public health, and social sciences. Existing statistical methods often rely on pre-specified models provided by experts, which can vary significantly, making causal results less robust. Other methods use causal discovery to estimate causal relationships from data, but the full structure cannot be learned. We examine how to combine these methods by learning parts of the structure from data and incorporating background knowledge to further restrict the graph structure.
Recent advancements have shown that it is possible to make causal inferences without knowing the full structure of the data-generating process, thus advancing causal discovery methods. We investigate the assumptions required to make precise causal inferences based on the graph structure and present methods for bounding causal effects when exact inference is not possible.

Vejledere: Niels Richard Hansen
Claus Ekstrøm, Anne Heldby Pedersen, SUND
Censor:     Klaus Kähler Holst, Mærsk