DeCaFlow: A Deconfounding Causal Generative Model
Published in Neural Information Processing Systems 2025 (NeurIPS 2025), San Diego, CA, USA (2-7 Dec 2025), 2025
Accepted as a Spotlight paper at NeurIPS 2025
A practical Causal Generative Model that allows for estimating interventional distributions and counterfactuals, even with unobserved confounders present. DeCaFlow leverages variational inference and Causal Normalizing Flows to model a Confounded Structural Causal Models, that yields in unbiased estimations of causal queries that are identifiable through adjustment and Proximal Inference.
The code implementation can be found in this link
Cite as:
@article{almodovar2025decaflow,
title={DeCaFlow: A Deconfounding Causal Generative Model},
author={Almod{\'o}var, Alejandro and Javaloy, Adri{\'a}n and Parras, Juan and Zazo, Santiago and Valera, Isabel},
journal={arXiv preprint arXiv:2503.15114},
year={2025}
}
}