Publication
(Psi)net: Efficient Causal Modeling at Scale
Florian Peter Busch; Moritz Willig; Jonas Seng; Kristian Kersting; Devendra Singh Dhami
In: Johan Kwisthout; Silja Renooij (Hrsg.). International Conference on Probabilistic Graphical Models, De Lindenberg, Nijmegen, the Netherlands, 11-13 September 2024. International Conference on Probabilistic Graphical Models (PGM), Pages 452-469, Proceedings of Machine Learning Research, Vol. 246, PMLR, 2024.
Abstract
Being a ubiquitous aspect of human cognition, causality has made its way into modern-day
machine-learning research. Despite its importance in real-world applications, contemporary
research still struggles with high-dimensional causal problems. Leveraging the efficiency of
probabilistic circuits, which offer tractable computation of marginal probabilities, we intro-
duce Ψnet, a probabilistic model designed for large-scale causal inference. Ψnet is a type
of sum-product network where layering and the einsum operation allow for efficient paral-
lelization. By incorporating interventional data into the learning process, the model can
learn the effects of interventions and make predictions based on the specific interventional
setting. Overall, Ψnet is a causal probabilistic circuit that efficiently answers causal queries
in large-scale problems. We present evaluations conducted on both synthetic data and a
substantial real-world dataset, demonstrating Ψnet’s ability to capture causal relationships
in high-dimensional settings.