A Hybrid Algorithm for Learning Causal Networks using Uncertain Experts’ Knowledge

Christophe Gonzales, Axel Journe, Ahmed Mabrouk
Proceedings of The 11th International Conference on Probabilistic Graphical Models, PMLR 186:241-252, 2022.

Abstract

Bayesian networks (BN) have become one of the most popular frameworks in causal studies. The causal relations between variables are encoded by the structure of the model, which is a directed acyclic graph (DAG). Unfortunately, despite the significant advances in algorithm development, learning the causal structure from data remains a very challenging task, especially for cases with a large number of variables. When the learning algorithm fails to identify the causal orientation of some edges, the human expert can provide some rough guidelines to complete the causal discovery. In many application domains, the expert knowledge might be uncertain about the right orientation of the edge. Worst, it may contradict the orientations learned from observational data, hence leading to conflicting situations. This paper presents a new hybrid algorithm combining a constraint-based approach with a greedy search, that includes specific rules to cope with uncertain domain/expert knowledge at different steps of the learning process. Experiments show the robustness of our method compared to other state-of-the-art algorithms.

Cite this Paper

BibTeX
@InProceedings{pmlr-v186-gonzales22a, title = {A Hybrid Algorithm for Learning Causal Networks using Uncertain Experts’ Knowledge}, author = {Gonzales, Christophe and Journe, Axel and Mabrouk, Ahmed}, booktitle = {Proceedings of The 11th International Conference on Probabilistic Graphical Models}, pages = {241--252}, year = {2022}, editor = {Salmerón, Antonio and Rumı́, Rafael}, volume = {186}, series = {Proceedings of Machine Learning Research}, month = {05--07 Oct}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v186/gonzales22a/gonzales22a.pdf}, url = {https://proceedings.mlr.press/v186/gonzales22a.html}, abstract = {Bayesian networks (BN) have become one of the most popular frameworks in causal studies. The causal relations between variables are encoded by the structure of the model, which is a directed acyclic graph (DAG). Unfortunately, despite the significant advances in algorithm development, learning the causal structure from data remains a very challenging task, especially for cases with a large number of variables. When the learning algorithm fails to identify the causal orientation of some edges, the human expert can provide some rough guidelines to complete the causal discovery. In many application domains, the expert knowledge might be uncertain about the right orientation of the edge. Worst, it may contradict the orientations learned from observational data, hence leading to conflicting situations. This paper presents a new hybrid algorithm combining a constraint-based approach with a greedy search, that includes specific rules to cope with uncertain domain/expert knowledge at different steps of the learning process. Experiments show the robustness of our method compared to other state-of-the-art algorithms.} }
Endnote
%0 Conference Paper %T A Hybrid Algorithm for Learning Causal Networks using Uncertain Experts’ Knowledge %A Christophe Gonzales %A Axel Journe %A Ahmed Mabrouk %B Proceedings of The 11th International Conference on Probabilistic Graphical Models %C Proceedings of Machine Learning Research %D 2022 %E Antonio Salmerón %E Rafael Rumı́ %F pmlr-v186-gonzales22a %I PMLR %P 241--252 %U https://proceedings.mlr.press/v186/gonzales22a.html %V 186 %X Bayesian networks (BN) have become one of the most popular frameworks in causal studies. The causal relations between variables are encoded by the structure of the model, which is a directed acyclic graph (DAG). Unfortunately, despite the significant advances in algorithm development, learning the causal structure from data remains a very challenging task, especially for cases with a large number of variables. When the learning algorithm fails to identify the causal orientation of some edges, the human expert can provide some rough guidelines to complete the causal discovery. In many application domains, the expert knowledge might be uncertain about the right orientation of the edge. Worst, it may contradict the orientations learned from observational data, hence leading to conflicting situations. This paper presents a new hybrid algorithm combining a constraint-based approach with a greedy search, that includes specific rules to cope with uncertain domain/expert knowledge at different steps of the learning process. Experiments show the robustness of our method compared to other state-of-the-art algorithms.
APA
Gonzales, C., Journe, A. & Mabrouk, A.. (2022). A Hybrid Algorithm for Learning Causal Networks using Uncertain Experts’ Knowledge. Proceedings of The 11th International Conference on Probabilistic Graphical Models, in Proceedings of Machine Learning Research 186:241-252 Available from https://proceedings.mlr.press/v186/gonzales22a.html.

Related Material