CEPAE: Conditional Entropy-Penalized Autoencoders for Time Series Counterfactuals

Tomàs Garriga [email protected], Gerard Sanz [email protected], Eduard Serrahima de Cambra [email protected], Axel Brando [email protected] · 2026

Read CEPAE: Conditional Entropy-Penalized Autoencoders for Time Series Counterfactuals by Tomàs Garriga [email protected], Gerard Sanz [email protected], Eduard Serrahima de Cambra [email protected], Axel Brando [email protected] (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-15546.txt.

Description

Garriga and colleagues build a tool for what-if questions about time series, where the aim is to judge how an event or a treatment shaped outcomes over time. Such questions matter in finance, health care, and marketing, but older counterfactual methods were not built for time series data. The team adapts variational and adversarial autoencoder methods to that setting, then goes further with a new design.

Their Conditional Entropy-Penalized Autoencoder, or CEPAE, adds an entropy penalty over the latent space to push learned features apart into cleaner parts. The method follows the abduction-action-prediction steps inside a structural causal model frame. Tests on synthetic, semi-synthetic, and real data show CEPAE beating the adapted methods on the measures used, with theory to back the design.

Catalog record

CREATORTomàs Garriga [email protected], Gerard Sanz [email protected], Eduard Serrahima de Cambra [email protected], Axel Brando [email protected]
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE92 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-15546

Browse the full library · About the Tower of Babel