A Novel Autoencoder Based Approach for Counterfactual Estimation Using Sparsity Constraints

Tomas Garriga, Gerard Sanz, Eduard Serrahima de Cambra, Axel Brando

Keywords: Counterfactuals, Causal Machine Learning, Causality, Time Series

Abstract: Building upon the abduction-action-step scheme and the structural causal model framework, this paper introduces the Conditional Sparse Autoencoder (CSAE), a novel approach for time series counterfactual estimation using encoder-decoder based architectures with a sparsity constraint to disentangle the roles of the inputs in the expected outputs. We benchmark CSAE with Conditional Variational Autoencoder (CVAE), the most widely adopted encoder-decoder architecture for counterfactual estimation, showing that CSAE clearly outperforms CVAE in this domain. Furthermore, we demonstrate the versatility of CSAE by extending it to image-based counterfactual scenarios, obtaining promising results. This work has important implications for a wide range of applications across various domains including finance, healthcare, and transportation, where being able to perform accurate counterfactual estimations is critical for decision-making.

Primary Area: causal reasoning

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 8088