DeCaFlow: A Deconfounding Causal Generative Model

Alejandro Almodóvar, Adrián Javaloy, Juan Parras, Santiago Zazo, Isabel Valera

Keywords: causal-inference, normalizing flows, tractable causal estimation

Abstract: We introduce DeCaFlow, a deconfounding causal generative model. In stark contrast to prior works, DeCaFlow requires training once per dataset with observational data and the causal graph, and enables accurate causal inference on continuous variables under the presence of hidden confounders. We extend previous theoretical results to show that a single instance of DeCaFlow provides correct estimates for all causal queries identifiable with do-calculus, leveraging proxy variables when do-calculus alone is insufficient. Moreover, we extend these results to counterfactual queries as well. Our empirical results on datasets such as Ecoli70—with 3 independent hidden confounders, tens of observed variables and hundreds of causal queries—show that DeCaFlow outperforms existing approaches, while demonstrating its out-of-the box applicability to any given causal graph.

Submission Number: 12