Causal Structure Learning Assumptions Shape Counterfactual Safety: Expert-Guided Constraints vs. Data-Driven DAGs with Probabilistic Logic Twin Networks
Entropy · 21.05.2026 · 28 (5) · 577
DOI: 10.3390/e28050577 ↗Kurz gefasst
Fachartikel in der Zeitschrift Entropy, online erschienen am 21.05.2026. Verfasst von Héctor Avilés und 11 weiteren Autor:innen. Beteiligte Einrichtung: Department of Information Technologies, Polytechnic University of Victoria, Ciudad Victoria 87138, Mexico. Die Arbeit nennt 57 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: UNAM-DGAPA, AI Consortium—CIMAT-CONAHCYT.
Abstract (Originalsprache)
We investigate how causal DAG learning algorithms and structural assumptions influence counterfactual decision safety. Four structure learning regimes are compared: expert-guided edge-constrained HC+BIC, unconstrained HC+BIC, MMPC+HC+BIC, and the PC-Stable algorithm. Evaluation is conducted using a leave-one-state-out protocol over a fully enumerated state–action space in a controlled offline autonomous driving setting. The environment is characterized by seven Boolean state variables and six actions, allowing us to disentangle the effects of learning strategies on counterfactual decisions. All models are implemented as probabilistic logic twin networks (PLTNs), with additional sensitivity analysis across parameter configurations. The learning regimes produce markedly different counterfactual decisions. Edge-constrained HC+BIC recommends a diverse set of safe actions, while unconstrained HC+BIC yields fewer but consistently safe alternatives. MMPC+HC+BIC frequently fails to identify safe actions, often associated with weak connectivity of the outcome variable. PC-Stable produces varied recommendations but may include unsafe actions, which is linked to incorrect edge orientations between actions and outcomes. These findings show that structure learning choices and prior knowledge influence counterfactual decisions through the learned structure, affecting the identification of safe alternatives in safety-critical applications.
Schlagwörter
Stichwörter der Autor:innen
Autor:innen und Institutionen
- Department of Information Technologies, Polytechnic University of Victoria, Ciudad Victoria 87138, Mexico.
- Institute of Computational Logic, Vienna University of Technology, 1040 Vienna, Austria.
- Computer Science Institute, Technological University of the Mixteca, Huajuapan de León 69004, Mexico.
- Control, Electronics and Communications Department, National Institute of Electricity and Clean Energies, Cuernavaca 62490, Mexico.
- Faculty of Engineering, National Autonomous University of Mexico, Mexico City 04510, Mexico.
- Center for Research and Advanced Studies, The National Polytechnic Institute, Tamaulipas Campus, Ciudad Victoria 87138, Mexico.
ORCID-Profile: Héctor Avilés, Ingridh Gracia, Verónica Rodríguez, Rubén Machucho, Alberto Reyes, Jesús Medrano, Felix Weitkämper
Förderung
- UNAM-DGAPA
- AI Consortium—CIMAT-CONAHCYT
Zitieren
APA
Avilés, H., Gracia, I., Kiesel, R., Rodríguez, V., Machucho, R., Reyes, A., Negrete, M., Ramírez, G., Luévano, N., Pequeño, M., Medrano, J., & Weitkämper, F. (2026). Causal Structure Learning Assumptions Shape Counterfactual Safety: Expert-Guided Constraints vs. Data-Driven DAGs with Probabilistic Logic Twin Networks. Entropy, 28(5), 577. https://doi.org/10.3390/e28050577
Vancouver
Avilés H, Gracia I, Kiesel R, Rodríguez V, Machucho R, Reyes A, et al. Causal Structure Learning Assumptions Shape Counterfactual Safety: Expert-Guided Constraints vs. Data-Driven DAGs with Probabilistic Logic Twin Networks. Entropy. 2026;28(5):577. doi:10.3390/e28050577
BibTeX
@article{Avils2026,
author = {Avilés, Héctor and Gracia, Ingridh and Kiesel, Rafael and Rodríguez, Verónica and Machucho, Rubén and Reyes, Alberto and Negrete, Marco and Ramírez, Gabriel and Luévano, Nicolás and Pequeño, Myriam and Medrano, Jesús and Weitkämper, Felix},
title = {Causal Structure Learning Assumptions Shape Counterfactual Safety: Expert-Guided Constraints vs. Data-Driven DAGs with Probabilistic Logic Twin Networks},
journal = {Entropy},
year = {2026},
volume = {28},
number = {5},
pages = {577},
doi = {10.3390/e28050577}
}
EUVIMED