The impact of prior knowledge on causal structure learning

Causal Bayesian Networks (CBNs) have become a powerful technology for reasoning under uncertainty, particularly in areas that require transparency and explainability, and rely on causal assumptions that enable us to simulate the effect of intervention. The graphical structure of these models can be estimated by causal knowledge, estimated from data using structure learning algorithms, or a combination of both. Various knowledge approaches have been proposed in the literature that enable us to specify prior knowledge that constrains or guides these algorithms. The objective of this paper is to investigate the impact of causal knowledge on structure learning across different settings that we might encounter in practice. We have achieved this by using a more comprehensive set of old and new knowledge approaches that enable us to obtain knowledge from heterogeneous sources, and considered a more comprehensive list of algorithms, case studies, and experimental settings. Each approach is assessed in terms of structure learning effectiveness and efficiency, including graphical accuracy, model fitting, complexity, and runtime; making this the first paper that provides a comparative evaluation of a wide range of knowledge approaches for structure learning. Because the value of knowledge depends on what data are available, we illustrate the results both with limited and big data. While the overall results show that knowledge becomes less important with big data due to higher learning accuracy rendering knowledge less important, some of the knowledge approaches are actually found to be more important with big data. Amongst the main conclusions is the observation that reduced search space obtained from knowledge does not always imply reduced computational complexity, perhaps because the relationships implied by the data and knowledge are in tension.