Causal normalizing flows: from theory to practice
Abstract
In this work, we deepen on the use of normalizing flows for causal reasoning. Specifically, we first leverage recent results on nonlinear ICA to show that causal models are identifiable from observational data given a causal ordering, and thus can be recovered using autoregressive normalizing flows (NFs). Second, we analyze different design and learning choices for causal normalizing flows to capture the underlying causal datagenerating process. Third, we describe how to implement the dooperator in causal NFs, and thus, how to answer interventional and counterfactual questions. Finally, in our experiments, we validate our design and training choices through a comprehensive ablation study; compare causal NFs to other approaches for approximating causal models; and empirically demonstrate that causal NFs can be used to address realworld problems, where the presence of mixed discretecontinuous data and partial knowledge on the causal graph is the norm. The code for this work can be found at https://github.com/psanch21/causalflows.
 Publication:

arXiv eprints
 Pub Date:
 June 2023
 DOI:
 10.48550/arXiv.2306.05415
 arXiv:
 arXiv:2306.05415
 Bibcode:
 2023arXiv230605415J
 Keywords:

 Computer Science  Machine Learning;
 Computer Science  Artificial Intelligence;
 Statistics  Methodology;
 Statistics  Machine Learning
 EPrint:
 31 pages, 15 figures. Under submission